Congressional Testimony
Here's a look at documents involving congressional testimony and member statements
Congressional Testimony
Featured Stories
Roxbury Institute Medical Director Amron Testifies Before Senate Special Committee on Aging
WASHINGTON, Aug. 19 -- The Senate Special Committee on Aging released the following testimony by David Amron, medical director of the Roxbury Institute, Los Angeles, California, and chair and founder of the Lipedema Society, from a July 29, 2026, hearing entitled "The AI Deception Machine: Deepfakes, Chatbots, and the New Frontier of Senior Fraud":
* * *
Chairman Scott, Ranking Member Gillibrand, and distinguished members of the Senate Special Committee on Aging, thank you for the opportunity to testify before you today.
My name is Dr. David Amron, and I am the Founder and Medical Director of ... Show Full Article WASHINGTON, Aug. 19 -- The Senate Special Committee on Aging released the following testimony by David Amron, medical director of the Roxbury Institute, Los Angeles, California, and chair and founder of the Lipedema Society, from a July 29, 2026, hearing entitled "The AI Deception Machine: Deepfakes, Chatbots, and the New Frontier of Senior Fraud": * * * Chairman Scott, Ranking Member Gillibrand, and distinguished members of the Senate Special Committee on Aging, thank you for the opportunity to testify before you today. My name is Dr. David Amron, and I am the Founder and Medical Director ofThe Roxbury Institute in Los Angeles, California, and the Founder and Chair of The Lipedema Society.
For more than three decades, I have dedicated my career to treating patients with lipedema, a chronic and often misunderstood disease that affects millions of women worldwide. I am widely recognized as the pioneer of lipedema surgery in North America and have devoted my career to advancing surgical treatment, educating physicians, and advocating on behalf of patients to help them obtain insurance coverage for their care.
As physicians, trust is the foundation of everything we do. Patients rely on us during some of the most vulnerable moments of their lives. I have spent more than three decades building that trust, yet scammers used artificial intelligence to exploit it in a matter of moments. Although I knew artificial intelligence was rapidly evolving, I never imagined how quickly it could be weaponized to deceive patients until it happened to me.
We first became aware of the scam in the summer of 2025, with reports increasing significantly by early September. Patients began contacting our office after seeing a video that appeared to show me endorsing a so-called "miracle" lipedema cream. Many forwarded us the video, but by then, some had already purchased the product before realizing it was fraudulent, including several of my own patients.
The video was disturbingly realistic. Scammers used actual footage from my YouTube channel and combined it with a fully AI-generated likeness and voice of my colleague, Dr. Karen Herbst, Head of Research and Director of Diagnostic and Preventative Medicine at The Roxbury Institute, as well as AI-generated celebrity images and stolen media logos, to create a polished advertisement that appeared entirely legitimate.
As soon as we recognized it as a coordinated deepfake, my team immediately issued public advisories across our social media platforms to warn patients, clarify that neither Dr. Herbst nor I had any involvement, and help our community recognize and report AI-generated medical impersonations.
Our web developer and digital marketing strategist immediately began investigating the scam as soon as patients alerted us. Before taking action, he investigated the company behind the product, Svelta Venastra, and quickly confirmed it was not a legitimate medical entity. He reported every version of the advertisement to Meta, the fraudulent accounts responsible for publishing it, the domain registrars hosting the websites, and the major search engines directing consumers to the scam. Yet each time one version was removed, another quickly appeared. This continued over the course of eleven days.
Despite these persistent efforts, the scam kept recurring. It quickly became clear that the existing reporting systems were not enough to stop a coordinated AI-enabled fraud campaign. After exhausting every avenue available to us, we realized we needed to bring national attention to what was happening. After exhausting every avenue available to us, we reached out to major television networks because we believed public awareness might accomplish what traditional reporting mechanisms could not. The TODAY Show responded, conducted its own investigation, and ultimately brought the scam to a national audience.
The TODAY Show attempted to identify and contact the individuals behind Svelta Venastra, just as we had. Like us, they found no legitimate company, no accountable representative, and no one willing to respond. Although the fraudulent video was eventually removed, versions of the scam have since resurfaced, underscoring how persistent and difficult these AI-enabled fraud schemes are to eliminate.
What disturbed me most was not that someone had misused my image. It was knowing that scammers had used my name and reputation to deceive the very people I have dedicated my career to helping.
The consequences extend far beyond financial loss. Patients may delay legitimate medical care for a progressive disease like lipedema, place their trust in unproven products and services, and allow their condition to worsen. Perhaps most damaging, these scams erode the trust patients place in their physicians. Once that trust is broken, patients no longer know whom they can trust.
Artificial intelligence has fundamentally changed the landscape of fraud, giving scammers the ability to convincingly impersonate trusted physicians, fabricate endorsements, and deceive patients on an unprecedented scale.
What happened to me is not an isolated incident. Physicians across the country are discovering that their images, voices, and professional reputations are being used without their knowledge or consent to promote products they have never evaluated, recommended, or even heard of.
The danger is clear. These scams put patients at risk and undermine the credibility of the medical profession. Older Americans are especially vulnerable because they often place tremendous trust in their physicians. Many also rely on the internet, including social media and online search results, to learn about medical conditions, making them especially susceptible when fraudulent content appears to come from trusted medical professionals.
When they see a realistic online video of a trusted physician recommending a treatment, they have little reason to question its authenticity. That is precisely what makes today's AI-generated deepfakes such a powerful tool for deception.
This problem is no longer theoretical. It is happening right now, and the consequences for patient safety, medical ethics, and the integrity of healthcare information are profound. If we do not act, more older Americans and other vulnerable patients will become victims of increasingly sophisticated AI-enabled fraud. I respectfully urge Congress to strengthen protections against AI-generated impersonation scams, improve accountability for those who create and distribute them, and ensure our laws keep pace with technology that can so easily be used to exploit trust for financial gain.
Thank you for the opportunity to testify today. I look forward to answering your questions.
* * *
Original text here: https://www.aging.senate.gov/imo/media/doc/32adb490-b498-1617-26d8-a24c018babde/Testimony_Amron%2007.29.26_0b05f233-2062-4676-b1ac-0a5a044873b8.pdf
* * *
Chairman Scott, Ranking Member Gillibrand, and distinguished members of the Senate Special Committee on Aging, thank you for the opportunity to testify before you today.
My name is Dr. David Amron, and I am the Founder and Medical Director of ... Show Full Article WASHINGTON, Aug. 19 -- The Senate Special Committee on Aging released the following testimony by David Amron, medical director of the Roxbury Institute, Los Angeles, California, and chair and founder of the Lipedema Society, from a July 29, 2026, hearing entitled "The AI Deception Machine: Deepfakes, Chatbots, and the New Frontier of Senior Fraud": * * * Chairman Scott, Ranking Member Gillibrand, and distinguished members of the Senate Special Committee on Aging, thank you for the opportunity to testify before you today. My name is Dr. David Amron, and I am the Founder and Medical Director ofThe Roxbury Institute in Los Angeles, California, and the Founder and Chair of The Lipedema Society.
For more than three decades, I have dedicated my career to treating patients with lipedema, a chronic and often misunderstood disease that affects millions of women worldwide. I am widely recognized as the pioneer of lipedema surgery in North America and have devoted my career to advancing surgical treatment, educating physicians, and advocating on behalf of patients to help them obtain insurance coverage for their care.
As physicians, trust is the foundation of everything we do. Patients rely on us during some of the most vulnerable moments of their lives. I have spent more than three decades building that trust, yet scammers used artificial intelligence to exploit it in a matter of moments. Although I knew artificial intelligence was rapidly evolving, I never imagined how quickly it could be weaponized to deceive patients until it happened to me.
We first became aware of the scam in the summer of 2025, with reports increasing significantly by early September. Patients began contacting our office after seeing a video that appeared to show me endorsing a so-called "miracle" lipedema cream. Many forwarded us the video, but by then, some had already purchased the product before realizing it was fraudulent, including several of my own patients.
The video was disturbingly realistic. Scammers used actual footage from my YouTube channel and combined it with a fully AI-generated likeness and voice of my colleague, Dr. Karen Herbst, Head of Research and Director of Diagnostic and Preventative Medicine at The Roxbury Institute, as well as AI-generated celebrity images and stolen media logos, to create a polished advertisement that appeared entirely legitimate.
As soon as we recognized it as a coordinated deepfake, my team immediately issued public advisories across our social media platforms to warn patients, clarify that neither Dr. Herbst nor I had any involvement, and help our community recognize and report AI-generated medical impersonations.
Our web developer and digital marketing strategist immediately began investigating the scam as soon as patients alerted us. Before taking action, he investigated the company behind the product, Svelta Venastra, and quickly confirmed it was not a legitimate medical entity. He reported every version of the advertisement to Meta, the fraudulent accounts responsible for publishing it, the domain registrars hosting the websites, and the major search engines directing consumers to the scam. Yet each time one version was removed, another quickly appeared. This continued over the course of eleven days.
Despite these persistent efforts, the scam kept recurring. It quickly became clear that the existing reporting systems were not enough to stop a coordinated AI-enabled fraud campaign. After exhausting every avenue available to us, we realized we needed to bring national attention to what was happening. After exhausting every avenue available to us, we reached out to major television networks because we believed public awareness might accomplish what traditional reporting mechanisms could not. The TODAY Show responded, conducted its own investigation, and ultimately brought the scam to a national audience.
The TODAY Show attempted to identify and contact the individuals behind Svelta Venastra, just as we had. Like us, they found no legitimate company, no accountable representative, and no one willing to respond. Although the fraudulent video was eventually removed, versions of the scam have since resurfaced, underscoring how persistent and difficult these AI-enabled fraud schemes are to eliminate.
What disturbed me most was not that someone had misused my image. It was knowing that scammers had used my name and reputation to deceive the very people I have dedicated my career to helping.
The consequences extend far beyond financial loss. Patients may delay legitimate medical care for a progressive disease like lipedema, place their trust in unproven products and services, and allow their condition to worsen. Perhaps most damaging, these scams erode the trust patients place in their physicians. Once that trust is broken, patients no longer know whom they can trust.
Artificial intelligence has fundamentally changed the landscape of fraud, giving scammers the ability to convincingly impersonate trusted physicians, fabricate endorsements, and deceive patients on an unprecedented scale.
What happened to me is not an isolated incident. Physicians across the country are discovering that their images, voices, and professional reputations are being used without their knowledge or consent to promote products they have never evaluated, recommended, or even heard of.
The danger is clear. These scams put patients at risk and undermine the credibility of the medical profession. Older Americans are especially vulnerable because they often place tremendous trust in their physicians. Many also rely on the internet, including social media and online search results, to learn about medical conditions, making them especially susceptible when fraudulent content appears to come from trusted medical professionals.
When they see a realistic online video of a trusted physician recommending a treatment, they have little reason to question its authenticity. That is precisely what makes today's AI-generated deepfakes such a powerful tool for deception.
This problem is no longer theoretical. It is happening right now, and the consequences for patient safety, medical ethics, and the integrity of healthcare information are profound. If we do not act, more older Americans and other vulnerable patients will become victims of increasingly sophisticated AI-enabled fraud. I respectfully urge Congress to strengthen protections against AI-generated impersonation scams, improve accountability for those who create and distribute them, and ensure our laws keep pace with technology that can so easily be used to exploit trust for financial gain.
Thank you for the opportunity to testify today. I look forward to answering your questions.
* * *
Original text here: https://www.aging.senate.gov/imo/media/doc/32adb490-b498-1617-26d8-a24c018babde/Testimony_Amron%2007.29.26_0b05f233-2062-4676-b1ac-0a5a044873b8.pdf
Martinez (Calif.) Victim of AI-Enabled Scam Testifies Before Senate Special Committee on Aging
WASHINGTON, Aug. 19 -- The Senate Special Committee on Aging released the following testimony by Deborah Del Mastro, victim of AI-enabled scam from Martinez, California, from a July 29, 2026, hearing entitled "The AI Deception Machine: Deepfakes, Chatbots, and the New Frontier of Senior Fraud":
* * *
The day started with a phone call at 10:22 from 925-453-0742 which had no name. I answered and said Hello and a male voice said "Hello? Who is this?" And I said "Who is this?" and he said I have your daughter here, does she have panic attacks? Then he put her on the phone with her crying and panicking ... Show Full Article WASHINGTON, Aug. 19 -- The Senate Special Committee on Aging released the following testimony by Deborah Del Mastro, victim of AI-enabled scam from Martinez, California, from a July 29, 2026, hearing entitled "The AI Deception Machine: Deepfakes, Chatbots, and the New Frontier of Senior Fraud": * * * The day started with a phone call at 10:22 from 925-453-0742 which had no name. I answered and said Hello and a male voice said "Hello? Who is this?" And I said "Who is this?" and he said I have your daughter here, does she have panic attacks? Then he put her on the phone with her crying and panickingsaying she was sorry Mom...IT WAS HER VOICE and then he got back on and said that he has 10 pounds of cocaine in his trunk and my daughter saw a deal go down and his boss told him to take her. Now he tells me that this is a Mexican drug cartel boss and he doesn't care who he kills, that he wants $20K for my daughter and if not she can pay with what's between her legs.... I'm at the breakfast table, incredulous with my jaw dropped and I quickly got dressed and mouthed to my husband as I left "she's been kidnapped!" and I left the house, still on the phone with this criminal.
Paul had no idea where I was going or what was going on, he could just tell that it was serious. The man was barking orders to me about having to have this money right away or his boss will do something with her, he told me to bring a phone charger or make sure my phone is charged because if we lose touch I won't see her again. He made demands for the next 5 1/2 hours, first to send $2000 by Western Union to Mexico (Maria Sanchez Gonzalez, Mexico, Mexico Federal District ) He wanted $2500 and I told him that I didn't have it -just $2000. He told me to go to Martinez Walmart and that's where I used Moneygram to send $2000. I took a pic and texted him the receipt and then he said there was a problem with it and I'd have to resend it. I told him I do not have the money to do this and he told me that there was an error and I'd be reimbursed. Still didn't have the money.
Then he wanted the other $500 right now and I had it in another bank account. Then I found that I had locked my keys in the car at Walmart and had to call my husband to pick me up and he joined the nightmare, his bank accounts were in play and he ran me around to Pittsburg Walmart to Martinez twice. We were on the phone with this awful man the entire time - with him listening in for every sound, asking where I was and what was going on if I was silent, threatening my family's lives and telling me that he can sell my daughter right now.
I tried Safeway for sending the $500 and couldn't but Walgreens worked. He told me to not talk during the time I was in the store and leave the phone on so he could hear everything that happened. I took a pic of the receipt and texted it to his phone, now a different number, all the time threatening me if I didn't hurry.
He was going to drop off Sarah first at the Martinez Safeway, then the Pittsburg Walmart, then Lucky's in Martinez and then again Pittsburg Walmart where we finally ended up finishing the scam when he said "we've released your daughter at the front entrance of the store, call her" We were at the Walmart, I jumped out to get her and she wasn't there. Panicked, I called her and she answered quietly that she was at work and what's going on? I broke down and was so relieved that she was alright and furious that we had been scammed out of all that money and there was nothing to be done about it but report it. All of the transactions were Cash within 10 minutes or Cash pickup anywhere in Mexico so we knew all the money was gone.
My husband sent $1700, after my $2500 and then when we were on the way to pick her up he demanded another $2000 which we did not have. He bargained us down to what we did have - $1200 and my husband, Paul Vincent Cicco, sent that too. All the while we had to be silent and on the phone with this guy. When he did speak he threatened harm to us and my daughter - saying that he had kids and he would kill everyone if his kids were at risk and his boss didn't care about anyone and would kill anyone who was in his way. He also talked many times about how he could sell Sarah for $20 thousand dollars or she could pay her way in other ways, hinting that she would be sex trafficked. But he was doing us a favor because we were respectful of him.
He also had me talk with Sarah again, after we sent the $1700, about 4 hours into the call and HER VOICE said very sadly " I'm so sorry Mom, I love you" After that another man's voice came on who was much more hard ass about everything - the enforcer - and threatened harm to us and Sarah if we didn't come up with more money, that's when we sent the last $1200.
It was the worst day of my life. And Paul's too. Some people are so broken.
* * *
Original text here: https://www.aging.senate.gov/imo/media/doc/32adb490-b498-1617-26d8-a24c018babde/Testimony_Del%20Mastro%2007.29.26_d1d3081d-ee04-4fdc-847b-e0b6195a26df.pdf
* * *
The day started with a phone call at 10:22 from 925-453-0742 which had no name. I answered and said Hello and a male voice said "Hello? Who is this?" And I said "Who is this?" and he said I have your daughter here, does she have panic attacks? Then he put her on the phone with her crying and panicking ... Show Full Article WASHINGTON, Aug. 19 -- The Senate Special Committee on Aging released the following testimony by Deborah Del Mastro, victim of AI-enabled scam from Martinez, California, from a July 29, 2026, hearing entitled "The AI Deception Machine: Deepfakes, Chatbots, and the New Frontier of Senior Fraud": * * * The day started with a phone call at 10:22 from 925-453-0742 which had no name. I answered and said Hello and a male voice said "Hello? Who is this?" And I said "Who is this?" and he said I have your daughter here, does she have panic attacks? Then he put her on the phone with her crying and panickingsaying she was sorry Mom...IT WAS HER VOICE and then he got back on and said that he has 10 pounds of cocaine in his trunk and my daughter saw a deal go down and his boss told him to take her. Now he tells me that this is a Mexican drug cartel boss and he doesn't care who he kills, that he wants $20K for my daughter and if not she can pay with what's between her legs.... I'm at the breakfast table, incredulous with my jaw dropped and I quickly got dressed and mouthed to my husband as I left "she's been kidnapped!" and I left the house, still on the phone with this criminal.
Paul had no idea where I was going or what was going on, he could just tell that it was serious. The man was barking orders to me about having to have this money right away or his boss will do something with her, he told me to bring a phone charger or make sure my phone is charged because if we lose touch I won't see her again. He made demands for the next 5 1/2 hours, first to send $2000 by Western Union to Mexico (Maria Sanchez Gonzalez, Mexico, Mexico Federal District ) He wanted $2500 and I told him that I didn't have it -just $2000. He told me to go to Martinez Walmart and that's where I used Moneygram to send $2000. I took a pic and texted him the receipt and then he said there was a problem with it and I'd have to resend it. I told him I do not have the money to do this and he told me that there was an error and I'd be reimbursed. Still didn't have the money.
Then he wanted the other $500 right now and I had it in another bank account. Then I found that I had locked my keys in the car at Walmart and had to call my husband to pick me up and he joined the nightmare, his bank accounts were in play and he ran me around to Pittsburg Walmart to Martinez twice. We were on the phone with this awful man the entire time - with him listening in for every sound, asking where I was and what was going on if I was silent, threatening my family's lives and telling me that he can sell my daughter right now.
I tried Safeway for sending the $500 and couldn't but Walgreens worked. He told me to not talk during the time I was in the store and leave the phone on so he could hear everything that happened. I took a pic of the receipt and texted it to his phone, now a different number, all the time threatening me if I didn't hurry.
He was going to drop off Sarah first at the Martinez Safeway, then the Pittsburg Walmart, then Lucky's in Martinez and then again Pittsburg Walmart where we finally ended up finishing the scam when he said "we've released your daughter at the front entrance of the store, call her" We were at the Walmart, I jumped out to get her and she wasn't there. Panicked, I called her and she answered quietly that she was at work and what's going on? I broke down and was so relieved that she was alright and furious that we had been scammed out of all that money and there was nothing to be done about it but report it. All of the transactions were Cash within 10 minutes or Cash pickup anywhere in Mexico so we knew all the money was gone.
My husband sent $1700, after my $2500 and then when we were on the way to pick her up he demanded another $2000 which we did not have. He bargained us down to what we did have - $1200 and my husband, Paul Vincent Cicco, sent that too. All the while we had to be silent and on the phone with this guy. When he did speak he threatened harm to us and my daughter - saying that he had kids and he would kill everyone if his kids were at risk and his boss didn't care about anyone and would kill anyone who was in his way. He also talked many times about how he could sell Sarah for $20 thousand dollars or she could pay her way in other ways, hinting that she would be sex trafficked. But he was doing us a favor because we were respectful of him.
He also had me talk with Sarah again, after we sent the $1700, about 4 hours into the call and HER VOICE said very sadly " I'm so sorry Mom, I love you" After that another man's voice came on who was much more hard ass about everything - the enforcer - and threatened harm to us and Sarah if we didn't come up with more money, that's when we sent the last $1200.
It was the worst day of my life. And Paul's too. Some people are so broken.
* * *
Original text here: https://www.aging.senate.gov/imo/media/doc/32adb490-b498-1617-26d8-a24c018babde/Testimony_Del%20Mastro%2007.29.26_d1d3081d-ee04-4fdc-847b-e0b6195a26df.pdf
Maria Empanada Owner Cantarovici Testifies Before Senate Small Business & Entrepreneurship Committee
WASHINGTON, Aug. 19 -- The Senate Small Business and Entrepreneurship Committee released the following testimony by Lorena Cantarovici, owner and founder of Maria Empanada, Denver, Colorado, from an Aug. 5, 2026, hearing entitled "The Next 250: Small Businesses Leading the Way":
* * *
Madam Chair Ernst, Ranking Member Markey, Senator Hickenlooper, and distinguished members of the Committee:
Thank you for inviting me today. My name is Lorena Cantarovici, founder of Maria Empanada. I was born and raised in Argentina by a single mother. For many years, life was terribly difficult... but my mother ... Show Full Article WASHINGTON, Aug. 19 -- The Senate Small Business and Entrepreneurship Committee released the following testimony by Lorena Cantarovici, owner and founder of Maria Empanada, Denver, Colorado, from an Aug. 5, 2026, hearing entitled "The Next 250: Small Businesses Leading the Way": * * * Madam Chair Ernst, Ranking Member Markey, Senator Hickenlooper, and distinguished members of the Committee: Thank you for inviting me today. My name is Lorena Cantarovici, founder of Maria Empanada. I was born and raised in Argentina by a single mother. For many years, life was terribly difficult... but my mothermoved mountains so I could be educated. I studied accounting and became a banker in Buenos Aires. I was proud to bring home a small, steady salary because my mother could not find work. But it was not enough. For a time, my mom and I were essentially homeless... I was an accountant with a job, but without a secure place to sleep. I grew tired of constant instability. I wanted the chance to build a future. So, with one backpack, three hundred dollars, and no English, I came to the United States.
Like many immigrants, my first job was in a restaurant. And I fell in love - with the energy, the people, and the feeling that every day began again. A restaurant is much more than food. It is a collection of systems. As a former bank auditor, I understood systems, and in the restaurant business, I found a new way to use that strength.
I learned how to make empanadas from my mother, Maria. Years later, I began making them in my home kitchen and selling them to friends. I took a class through an SBA-supported program to learn how to write a business plan and turn an idea into a company. I borrowed four thousand dollars and opened my first location.... In that moment, I felt that perhaps the American Dream was not only something I had heard about. Perhaps it could belong to me, too.
Then, I did what entrepreneurs are told to do: I went to the banks. Twelve banks rejected me. I was considered "un-bankable." One banker joked, "We like lending money to people who do not need it." It turns out that was not really a joke. That was the root of the problem.
Eventually, community lenders gave me a chance. Then came an SBA loan, and later venture capital. In 2017, I became only the thirteenth minority woman in the country, in history, to receive an investment of that size. Now, after years of pouring myself into Maria Empanada, we have expanded to seven locations, including the Denver International Airport, and employ over fifty people in Colorado. The business does much more than just providing food to people - we provide community, culture, and a place to connect.
And my success has provided the privilege to talk to you today about the barriers that still exist for the many entrepreneurs that are hoping to walk down this path. The first is access to capital.
Only one-half of one percent of venture capital goes to minority women. Being one of the few is an honor. Remaining one of the few is a problem.
Another barrier is the current approach to immigration. Immigration is complex, difficult, and emotional. As a small business owner, I'll leave immigration policy to you all. But I do want to paint you a picture of how immigration policy affects businesses on the ground. Fear has entered everyday life for many people. People are afraid to go out because of their skin color, their face, their accent, or because they may be mistaken for someone else. Guests stay home instead of dining out. Employees miss work or avoid applying for jobs. People make themselves smaller and communities shrink. This fear affects people regardless of their legal status.
I am a new American, but I will always be an immigrant. There is an advantage to seeing this country through immigrant eyes: sometimes, you must come from somewhere else to appreciate how extraordinary its opportunities truly are. This country gave me the opportunity to become an entrepreneur. It gave me the opportunity to create something of my own, something for my community. But entrepreneurs do not only create companies. We create other entrepreneurs. Success is contagious. Someone sees what we built and begins to believe: "Maybe I can do it too."
Restaurants are struggling and they are a small piece of a much larger ecosystem that includes farmers, ranchers, drivers, suppliers, builders, customers and many others.
Restaurants are where people laugh, gather, and celebrate. We bring life and joy to neighborhoods and cities. I came to this country looking for better possibilities. As an entrepreneur, this country gave them to me. I hope together we can create a future where other entrepreneurs can do the same.
Thank you.
* * *
Original text here: https://www.sbc.senate.gov/public/_cache/files/8/e/8edaeea2-a923-4c7b-b586-996676ad0bc5/3BCD15E73987F40BA7CF7BBA23C9F079DA365BBB577C571F69B7ECCA1BA76006.cantarovici-testimony.pdf
* * *
Madam Chair Ernst, Ranking Member Markey, Senator Hickenlooper, and distinguished members of the Committee:
Thank you for inviting me today. My name is Lorena Cantarovici, founder of Maria Empanada. I was born and raised in Argentina by a single mother. For many years, life was terribly difficult... but my mother ... Show Full Article WASHINGTON, Aug. 19 -- The Senate Small Business and Entrepreneurship Committee released the following testimony by Lorena Cantarovici, owner and founder of Maria Empanada, Denver, Colorado, from an Aug. 5, 2026, hearing entitled "The Next 250: Small Businesses Leading the Way": * * * Madam Chair Ernst, Ranking Member Markey, Senator Hickenlooper, and distinguished members of the Committee: Thank you for inviting me today. My name is Lorena Cantarovici, founder of Maria Empanada. I was born and raised in Argentina by a single mother. For many years, life was terribly difficult... but my mothermoved mountains so I could be educated. I studied accounting and became a banker in Buenos Aires. I was proud to bring home a small, steady salary because my mother could not find work. But it was not enough. For a time, my mom and I were essentially homeless... I was an accountant with a job, but without a secure place to sleep. I grew tired of constant instability. I wanted the chance to build a future. So, with one backpack, three hundred dollars, and no English, I came to the United States.
Like many immigrants, my first job was in a restaurant. And I fell in love - with the energy, the people, and the feeling that every day began again. A restaurant is much more than food. It is a collection of systems. As a former bank auditor, I understood systems, and in the restaurant business, I found a new way to use that strength.
I learned how to make empanadas from my mother, Maria. Years later, I began making them in my home kitchen and selling them to friends. I took a class through an SBA-supported program to learn how to write a business plan and turn an idea into a company. I borrowed four thousand dollars and opened my first location.... In that moment, I felt that perhaps the American Dream was not only something I had heard about. Perhaps it could belong to me, too.
Then, I did what entrepreneurs are told to do: I went to the banks. Twelve banks rejected me. I was considered "un-bankable." One banker joked, "We like lending money to people who do not need it." It turns out that was not really a joke. That was the root of the problem.
Eventually, community lenders gave me a chance. Then came an SBA loan, and later venture capital. In 2017, I became only the thirteenth minority woman in the country, in history, to receive an investment of that size. Now, after years of pouring myself into Maria Empanada, we have expanded to seven locations, including the Denver International Airport, and employ over fifty people in Colorado. The business does much more than just providing food to people - we provide community, culture, and a place to connect.
And my success has provided the privilege to talk to you today about the barriers that still exist for the many entrepreneurs that are hoping to walk down this path. The first is access to capital.
Only one-half of one percent of venture capital goes to minority women. Being one of the few is an honor. Remaining one of the few is a problem.
Another barrier is the current approach to immigration. Immigration is complex, difficult, and emotional. As a small business owner, I'll leave immigration policy to you all. But I do want to paint you a picture of how immigration policy affects businesses on the ground. Fear has entered everyday life for many people. People are afraid to go out because of their skin color, their face, their accent, or because they may be mistaken for someone else. Guests stay home instead of dining out. Employees miss work or avoid applying for jobs. People make themselves smaller and communities shrink. This fear affects people regardless of their legal status.
I am a new American, but I will always be an immigrant. There is an advantage to seeing this country through immigrant eyes: sometimes, you must come from somewhere else to appreciate how extraordinary its opportunities truly are. This country gave me the opportunity to become an entrepreneur. It gave me the opportunity to create something of my own, something for my community. But entrepreneurs do not only create companies. We create other entrepreneurs. Success is contagious. Someone sees what we built and begins to believe: "Maybe I can do it too."
Restaurants are struggling and they are a small piece of a much larger ecosystem that includes farmers, ranchers, drivers, suppliers, builders, customers and many others.
Restaurants are where people laugh, gather, and celebrate. We bring life and joy to neighborhoods and cities. I came to this country looking for better possibilities. As an entrepreneur, this country gave them to me. I hope together we can create a future where other entrepreneurs can do the same.
Thank you.
* * *
Original text here: https://www.sbc.senate.gov/public/_cache/files/8/e/8edaeea2-a923-4c7b-b586-996676ad0bc5/3BCD15E73987F40BA7CF7BBA23C9F079DA365BBB577C571F69B7ECCA1BA76006.cantarovici-testimony.pdf
Indiana Business Research Center Director Rogers Testifies Before Senate Health, Education, Labor & Pensions Subcommittee
WASHINGTON, Aug. 19 -- The Senate Health, Education, Labor and Pensions Subcommittee on Employment and Workplace Safety released the following written testimony by Carol O. Rogers, director of the Indiana Business Research Center at the Indiana University Kelley School of Business, from a July 29, 2026, hearing entitled "The Impact of AI on the Workforce":
* * *
I. Introduction and Qualifications
Chairman Banks, Ranking Member Hickenlooper, and Members of the Subcommittee, thank you for the opportunity to submit this written statement in connection with the Subcommittee's hearing, "The Impact ... Show Full Article WASHINGTON, Aug. 19 -- The Senate Health, Education, Labor and Pensions Subcommittee on Employment and Workplace Safety released the following written testimony by Carol O. Rogers, director of the Indiana Business Research Center at the Indiana University Kelley School of Business, from a July 29, 2026, hearing entitled "The Impact of AI on the Workforce": * * * I. Introduction and Qualifications Chairman Banks, Ranking Member Hickenlooper, and Members of the Subcommittee, thank you for the opportunity to submit this written statement in connection with the Subcommittee's hearing, "The Impactof AI on the Workforce," and specifically on the AI Workforce PREPARE Act (S. 3339).[11]
I am Carol O. Rogers, Director of the Indiana Business Research Center (IBRC) at the Indiana University Kelley School of Business. I am here today as a subject matter expert; the views I express are my own and do not represent my employer, Indiana University.
As Director of IBRC, I work on issues related to Indiana's population, economy, and workforce to help Hoosier policymakers, educators, employers, and workers. In that role, my staff and I work daily with federal statistical series, including data from the Bureaus of Labor Statistics, Census, and Economic Analysis, as well as with Indiana's state administrative records, including unemployment insurance wage records and the state's longitudinal education-to-workforce data system. I have also served on the boards of several national research organizations, including as chairman of national organizations including the Labor Market Information Institute and the Council for Community and Economic Research.
I offer this testimony not as an advocate for a particular outcome on AI policy, but as a researcher with direct, practical experience in the strengths and limitations of the federal and state data infrastructure this bill would rely on. My comments focus on the available evidence showing, as best it can, AI's impact on the workforce, and on whether and where the underlying data infrastructure is ready to support this bill's goals.
II. What We Are Seeing: AI's Impact on the Workforce So Far
Generative and agentic AI has evolved quickly since ChatGPT's public release in late 2022.
Adoption has moved from early experimentation to routine business use: nearly 20 percent of U.S. firms now report using AI in at least some business function within the past two weeks (based on the latest data from February 2026), according to the U.S. Census Bureau's Business Trends and Outlook Survey.[1] Individual adoption has grown even faster. Harvard's Project on Workforce finds that 62 percent of U.S. adults ages 18 to 64 now use generative AI, and that 6.3 percent of total work hours are spent using it,[2] building on earlier academic research documenting the pace of generative AI's diffusion into daily work and jobs.[3]
* * *
Figure 1. Share of firms currently using AI in some business function, by state, February 2026.
Source: U.S. Census Bureau, Business Trends and Outlook Survey (BTOS), census.gov/hfp/btos.
* * *
This state-level detail from BTOS, shown in Figure 1, illustrates the unevenness of AI adoption across the country. Current AI use among firms ranges from a low of 11.1 percent in Arkansas to a high of 28.8 percent in Massachusetts, a gap of more than seventeen percentage points. Idaho (28.1 percent), Virginia (27.6 percent), Nevada (27.3 percent), and Washington (26.2 percent) round out the highest-adoption states, while Mississippi (11.8 percent), Nebraska (12.3 percent), New Mexico (13.6 percent), and Alabama (14.3 percent) report the lowest current use. A few patterns stand out. Adoption tends to run highest in states with large technology and professional-services sectors, such as Massachusetts, Virginia, and Washington, or with unusually concentrated industry clusters, as in Idaho and Nevada, and lowest in states with economies weighted toward agriculture and traditional manufacturing, concentrated in the Great Plains and Deep South. Neighboring states can also diverge sharply: Virginia's 27.6 percent sits beside West Virginia's 16.1 percent, and Idaho's 28.1 percent borders Oregon's 18.9 percent, a reminder that state averages can mask very different local economies just across a state line.
Indiana, at 20.9 percent, sits close to the roughly 20 percent national rate cited above, indicating that Indiana's experience is broadly representative of the national middle rather than an outlier in either direction.
My Center recently published research from the Indiana Department of Workforce (In Context, Mar-Apr 2026) comparing occupations with high and low AI exposure.[4] That analysis found job postings in AI-exposed occupations in Indiana shrank 41 percent since 2022, compared to a 35 percent decline in less-exposed occupations. Advertised wages in exposed occupations grew faster over the same period, 16 percent versus 11 percent, and the unemployment-rate advantage AI-exposed workers once held, a full percentage point in 2022, had nearly disappeared by last fall. Taken together, this looks less like mass job elimination and more like market-based repricing: some occupations losing ground, others gaining pay, often within the same broad category.
This pattern is consistent with what we are seeing nationally. The Federal Reserve Bank of Atlanta's survey of more than 700 corporate executives found little near-term aggregate employment decline attributable to AI, alongside a compositional reallocation of labor, with administrative support roles declining as demand grows for skilled technical roles.[5] The Federal Reserve Bank of Dallas has documented a similar wage pattern: nominal wages are up 7.5 percent across all industries since late 2022, but 8.5 percent in the most AI-exposed industries, and nearly 17 percent in computer systems design specifically.[6] Federal Reserve Governor Michael Barr summarized this dynamic in a February 2026 address, noting that AI's long-run effects on the labor market are likely to be positive, but that in the short term it can meaningfully disrupt specific workers, making how we manage the transition consequential.[7] At the same time, I would caution everyone against inferring too much from Amazon, Meta, and other major technology firms that have announced substantial AI-linked workforce reductions over the past year. Outplacement firm Challenger, Gray & Christmas reports that AI has been cited in roughly a fifth to a quarter of all announced U.S. layoffs so far in 2026.[8] But that pattern is concentrated in a narrow slice of large, capital-intensive technology firms making enormous, simultaneous investments in AI infrastructure. It is not what we see broadly across small and medium-sized businesses. The National Federation of Independent Business's most recent survey finds that 98 percent of small employers currently using AI report no change in their employee counts, and NFIB's monthly economic trends data show small business hiring plans holding at a modest net positive.[9] Policy built primarily around the experience of the largest companies risks missing what is actually happening for the overwhelming majority of American employers.
Finally, a broader macro signal deserves attention. Goods-producing employment declined over the course of last year and has largely plateaued in 2026, even as productivity and wages have continued to rise.[10] Fewer workers producing more value, at higher pay, is consistent with a meaningful contribution from automation and AI, working alongside other factors such as capital investment, though this causal link is not yet provable with the data currently available. It is exactly the kind of signal this bill's data infrastructure should help us track and understand with more precision than we can today.
III. Why Data Infrastructure Will Determine This Bill's Success
The AI Workforce PREPARE Act is, at its core, a data bill. Its central premise, stated plainly in its findings, is that policymakers, training providers, and workers currently lack the data and forecasts needed to anticipate and mitigate AI-driven worker dislocation.[11] I agree with that premise, and the evidence summarized in Section II above illustrates why: the picture is real, but it is also mixed, still emerging, and in places contradictory. Where I would like to focus the Subcommittee's attention is a second-order question the bill does not fully resolve: whether the federal and state data systems it directs agencies to use and improve are themselves currently capable of producing timely, reliable, AI-attributable signals about the labor market.
In my experience, the answer is mixed. Some of the infrastructure this bill would lean on (particularly state administrative wage records), is more capable than is generally understood in federal policy discussions, but underutilized. Other pieces, particularly the timeliness of occupational classification systems and the standardization of employer self-reported data, are not yet fit for the purpose this bill assigns them. Both of these points matter for implementation, funding, and oversight.
IV. Section-by-Section Assessment
A. AI Workforce Research Hub
The bill's creation of an AI Workforce Research Hub to convene researchers, employers, and labor, and to help implement the White House AI Action Plan, is a sound organizing structure.
Its value will depend heavily on whether it is designed to integrate existing state and federal data assets rather than to build a new, parallel data collection effort from scratch. States like Indiana have already made substantial investments in longitudinal data infrastructure; a federal hub that does not connect to this work risks duplicating effort and producing a less complete picture than what already exists in pieces across federal agencies and states.
B. WARN Act Amendment -- The "Substantial Factor" Standard
Amending the WARN Act to require disclosure when AI is a "substantial factor" in a mass layoff is a meaningful transparency measure, and I understand its intent. However, as a federal data consumer, I want to flag a specific risk: the bill directs the Secretary of Labor to issue guidance on how employers may determine that AI is a substantial factor and estimate the associated share of job loss, but employer compliance is explicitly satisfied by a "good-faith statement," without a standardized measurement method attached to it.[11]
Absent a common framework, I would expect significant variation across employers and industries in how "substantial factor" is interpreted, not necessarily out of bad faith, but because there is currently no agreed-upon, operational definition connecting a specific business decision to a specific automation event. This matters because data that cannot be standardized across employers cannot be aggregated into a reliable and comparable national and regional picture, which undercuts the purpose of collecting it and could lead to thousands of differing interpretations.
Recommendation: DOL's guidance should point employers toward existing, standardized classification systems, such as the Standard Occupational Classification (SOC) system and O*NET tasks and skills data, as a reference framework for these determinations, so that reported information is at least structurally comparable across firms, sectors, and states, and can be crosschecked against independent administrative data and other sources over time rather than resting solely on self-reporting.
C. BLS Occupational Forecasts for AI-Sensitive Occupations
BLS occupational projections, and the Standard Occupational Classification system that underpins them, update on multi-year cycles that are considerably slower than the pace at which AI is changing specific tasks and skills within occupations. An occupation can appear stable in the aggregate data long after AI has meaningfully altered the tasks performed within it and the skills required by the worker. I would encourage the Subcommittee to direct BLS toward task-level analysis, for example building on the O*NET task and skills content with more frequent employer surveys, rather than relying solely on occupation-level projections, which are a blunter instrument for this specific policy question.
In terms of resources, in my experience with comparable state and federal data modernization efforts, the anticipated funding over a five-year period is unlikely to be sufficient to meaningfully accelerate BLS's forecasting capacity for such a fast-moving technology. I would recommend the Subcommittee revisit this authorization level as implementation proceeds and consider it a floor rather than an adequate ceiling.
D. Researcher Access to Federal Workforce Data
The bill's provision to increase researchers' access to federal workforce data is, in my assessment, one of its most consequential and most under-discussed elements. Speaking from direct experience navigating Federal Statistical Research Data Center (FSRDC) access, restricted-use microdata agreements, and interagency data-sharing arrangements, I can attest that current approval timelines, inconsistent disclosure review standards across agencies, and limited capacity for linking federal data with state administrative records are all real, practical barriers to timely research, not merely inconveniences.
I would encourage the Subcommittee to seek specificity in how this provision is implemented: whether it means expanded FSRDC access, streamlined restricted-use data agreements, funding for state-federal data linkage infrastructure, or some combination. Without that specificity, this provision risks remaining a statement of intent rather than a change we can actually use.
E. State Administrative Records and In-Demand Occupation Lists
This is the area where I can speak most directly from operational experience, and where I believe the Subcommittee has the greatest opportunity to strengthen the bill.
States' unemployment insurance wage records, together with Statewide Longitudinal Data Systems (SLDS) that link education and workforce outcomes, already capture actual, verified employment and earnings changes on a quarterly basis, at a level of industry and geographic granularity that federal survey instruments cannot match. These records are, in effect, one of the most complete real-time pictures of the American labor market already in existence, and they are collected and maintained by every state, largely independent of federal AI-specific policy. In Indiana, we use these records daily to inform workforce and education policy, and I believe they may be underappreciated in Washington relative to their value.
Two considerations follow. First, I would recommend the bill explicitly direct the AI Workforce Research Hub and related data-collection provisions to connect with, rather than duplicate, this existing state infrastructure, including funding mechanisms for states to modernize and link these records, building on ongoing Enhanced Wage Records initiatives, for AI-workforce intelligence purposes. Second, I want to be candid that data definitions, industry coding practices, and system capacity vary meaningfully from state to state. This is a real and, I believe, solvable technical barrier, but it will require dedicated federal investment and technical assistance to states, not simply a data-sharing mandate, to produce comparable results across states, regions, and even congressional districts in ways that truly inform policy.
Finally, I want to flag the stakes of getting this right: the bill ties AI-informed labor market projections to states' in-demand occupation lists, which in turn determine which training programs receive federal workforce development funding under WIOA and related authorities.
This creates a direct chain from federal AI-workforce data quality to aligning which training program an individual worker is funded to attend. Forecast error at the federal level does not stay abstract; it becomes a misallocated training dollar and, for an individual worker, a missed opportunity.
F. Voluntary Public-Private Data Sharing and Prize Competitions
The bill's approach to facilitating voluntary data sharing from AI developers and running prize competitions to better understand AI adoption and task-level automation is a reasonable, low-friction starting point, particularly relative to more prescriptive mandatory-disclosure proposals also pending before Congress. My caution here is methodological: voluntarily submitted adoption data, whether from employers or AI companies, is likely to be subject to selection bias, both in which firms choose to participate and in how they characterize their own AI use. I would recommend that any resulting datasets be explicitly validated against independent administrative benchmarks, such as those described in Section IV.E, rather than treated as authoritative on their own.
V. My Experience in Indiana
My work at IBRC sits directly at the intersection of federal statistics and Indiana's own administrative records, and demand across the state for actionable intelligence on AI's impact is high. At the same time, standardized, complete, current, local data remains hard to come by.
Businesses ask us for competitive intelligence. Workers want career and training guidance in an AI-uncertain world. Policymakers want to know whether Indiana will have the workforce employers need. Educators need to align curricula.
We are estimating impact now using AI-exposure metrics by Indiana county, but we still do not know how employers are actually implementing AI, governing its use, integrating it into individual job tasks, or what specific AI skills they are hiring for, training toward, or planning around, the detail we need to gauge impact accurately. This is precisely the kind of employer-level, task-level information the researcher-access and data-hub provisions of this bill could help supply, if implemented with the state-data linkages described in Section IV.E above.
I do not come before this Subcommittee as an advocate for a predetermined conclusion about AI's impact, but as someone whose job it is to make sure Indiana's workers, educators, and policymakers know what is happening, and what is likely to happen, to our workforce.
VI. Recommendations
In summary, I offer the following recommendations for the Subcommittee's consideration as it advances this bill:
* * *
1. Direct BLS to prioritize task- and skill-level analysis, building on O*NET, alongside occupation-level projections, given the pace at which AI is changing work within occupations faster than occupational classifications themselves change.
* * *
2. Require DOL's WARN Act guidance to anchor employer "substantial factor" determinations to standardized occupational, skill, and task classifications, so reported data can be aggregated and compared across employers and states.
3. Revisit the funding authorization for BLS occupational forecasting as a floor, not a sufficient level of investment, given the scope of the task.
4. Specify what expanded researcher access to federal workforce data will concretely mean in practice, whether FSRDC capacity, restricted-use data agreement timelines, or state-federal linkage funding, rather than leaving the provision as a general statement of intent.
5. Explicitly direct the AI Workforce Research Hub and related provisions to connect with existing state administrative data infrastructure, including UI wage records as part of ongoing Enhanced Wage Records initiatives and Statewide Longitudinal Data Systems, and fund state-level technical assistance to improve cross-state comparability, rather than building new, parallel federal data collection.
6. Require independent validation of voluntarily submitted AI-adoption data against administrative benchmarks before it informs funding-linked determinations such as states' in-demand occupation lists.
7. Direct the AI Workforce Research Hub to develop a standardized instrument for capturing employer-level AI implementation, governance, and skills-demand data, since neither federal nor state administrative systems currently capture this information at the granularity needed to guide training investment.
VII. Conclusion
The bill reflects a sound, bipartisan recognition that this country needs better data to manage the workforce transitions AI is already producing. My purpose today has been to bring a research and data-infrastructure perspective to that effort: to describe what the available evidence currently shows, to highlight where existing federal and state data systems are ready to meet this bill's ambitions, and to flag where targeted investment and design choices will determine whether they can. I would welcome the opportunity to work with the Subcommittee and its staff as this legislation moves forward, and I am happy to answer any questions.
* * *
References
[1] U.S. Census Bureau. Business Trends and Outlook Survey (BTOS). 2026. census.gov/hfp/btos.
[2] Bick, Alexander, and Adam Blandin. GenAI Adoption Tracker. Data as of May 2026. genaiadoptiontracker.com.
[3] Bick, Alexander, Adam Blandin, and David J. Deming. "The Rapid Adoption of Generative AI." NBER Working Paper No. 32966. National Bureau of Economic Research, 2024 (revised 2026).
[4] InContext, "Is AI Affecting Indiana's Labor Market?" March-April 2026. Indiana Business Research Center, incontext.indiana.edu.
[5] Federal Reserve Bank of Atlanta. "How Might AI Change the Workplace? Evidence from Corporate Executives." Policy Hub, March 25, 2026; see also Working Paper 202604, "Artificial Intelligence, Productivity, and the Workforce: Evidence from Corporate Executives." atlantafed.org.
[6] Davis, J. Scott. "AI Is Simultaneously Aiding and Replacing Workers, Wage Data Suggest." Federal Reserve Bank of Dallas, February 24, 2026. dallasfed.org/research/economics/2026/0224.
[7] Barr, Michael S. "What Will Artificial Intelligence Mean for the Labor Market and the Economy?" Speech at the New York Association for Business Economics, February 17, 2026. federalreserve.gov.
[8] Challenger, Gray & Christmas, Inc. Monthly Job Cuts Report, 2026 (AI cited in approximately 22-23 percent of announced U.S. layoffs year-to-date as of mid-2026). challengergray.com.
[9] National Federation of Independent Business. "Small Business and Technology Survey." June 25, 2025; and NFIB Small Business Economic Trends (SBET) monthly reports, 2026. nfib.com.
[10]U.S. Bureau of Labor Statistics. Employment Situation and goods-producing employment series (CES), 2025-2026, retrieved via FRED, Federal Reserve Bank of St. Louis. fred.stlouisfed.org.
[11] AI Workforce PREPARE Act, S. 3339, 119th Congress (2025-2026). congress.gov/bill/119th-congress/senate-bill/3339.
* * *
Supplementary material appears below that is part of an Indiana Business Research Center working paper looking at AI exposure throughout Indiana, by county. I provide this as further insight into the level of detail sought in my state to understand the breadth of AI potential for employers and workers, as well as noting the potential risks.
Generative AI Exposure Assessment: Indiana Counties
An IBRC Workforce and Economic Development Analysis Working Paper
Indiana Business Research Center, Indiana University Kelley School of Business
* * *
Figure. Intensity of AI-exposed jobs across Indiana counties.
Source: IN Context, Mar-Apr 2026, Indiana Business Research Center, incontext.indiana.edu.
* * *
The figure above shows the county-level detail behind an estimate of the exposure to Indiana's occupations. Exposure scores cluster tightly across Indiana's 92 counties, ranging from roughly 52 at the low end to 60 at the high end, an eight-point spread, with a median near 56. Most counties fall within a narrow band close to that median, while a handful of counties scattered across both the northern and southern parts of the state register the highest scores on the map, and a single county in the north-central part of the state registers the lowest. This tight, dispersed clustering suggests AI exposure in Indiana is not confined to the state's largest metropolitan counties, but runs at meaningfully similar intensity across metropolitan, small-city, and rural counties alike, underscoring why sub-state, county-level data, and not just statewide or occupational averages, is necessary to guide training investment and workforce policy.
Methodology
Data Integration:
* Lightcast(TM) Employment Data: 2.7 million Indiana jobs across 909 industries.
* County Staffing Patterns: Occupation-industry matrices for each county.
* Felten AI Scores: Generative AI exposure by occupation and industry.
Key Innovation: Rather than applying generic industry scores, this analysis accounts for countyspecific occupational compositions. The same industry can have different AI exposure across counties based on local staffing patterns.
Two types of AI scores are combined to produce a composite score for each county and industry:
* Image Generation AI Exposure: measures how susceptible occupations are to being augmented or replaced by AI systems that create visual content, such as DALL-E, Midjourney, or Stable Diffusion. High scores indicate jobs where workers currently spend significant time on tasks like graphic design, illustration, photo editing, or visual content creation that AI image generators can now perform.
* Language Modeling AI Exposure: measures exposure to AI systems that understand and generate text, such as ChatGPT, Claude, or GPT-5. High scores indicate occupations involving substantial writing, analysis, research, communication, or information-processing tasks that large language models can augment or automate.
The two scores are combined into a composite AI score, an employment-weighted average of the two types of scores. The full calculation flow:
1. Occupation level: each occupation has an image-generation score and a language-modeling score, on a 0-100 scale.
2. Industry level: for each county-industry combination, each occupation's AI scores are weighted by its employment share within that industry, and the two AI types are averaged.
3. County level: each industry's composite score is weighted by its employment share within the county, producing a final, employment-weighted county score.
These scores measure AI risk: these scores focus specifically on generative AI capabilities that have rapidly advanced over the past two to three years, rather than on traditional automation or robotics. The underlying Felten et al. research identifies which job tasks are most exposed to these newer AI capabilities, providing a more current and relevant assessment of workforce disruption potential than older automation studies. Complementary coverage: image generation primarily affects creative, design, and visual-communication roles, while language modeling affects knowledge work, analysis, and text-based tasks. Together, they capture the breadth of generative AI's current capabilities and provide a fuller view of which occupations face the most immediate AI-driven change.
Coverage: 92 Indiana counties, representing 2,790,788 jobs.
Statewide Patterns
There is a strong positive relationship between counties with high language-modeling exposure and high image-generation exposure.
* Most of the low-exposure counties are rural, with the exception of Boone County, which is located in the Indianapolis metropolitan statistical area.
* There are fewer clear geographic patterns among the high-exposure counties, though most are part of metropolitan areas.
* Martin County stands out as the county with the highest AI exposure score in the state.
This small county's high score is driven by its outsized employment share, approximately 29 percent, in Engineering Services (NAICS 541330), a location quotient of 8.5, an industry with a high image-generation exposure score based on its local staffing patterns.
* Ohio County presents a useful counterexample. Given its high employment in hospitality, driven by the Rising Sun Casino Resort, it was surprising that this county also scored highly on AI exposure. This underlines an important caution for the Subcommittee: these exposure scores should not be relied on blindly for workforce planning. Local knowledge remains essential to interpreting them correctly.
Combining exposure level with county employment size points to a practical, tiered approach for workforce-development priority:
Priority Level ... Exposure Tier ... Employment Size ... Counties
Monitor ... Low Exposure ... Small ... 34
Monitor ... Moderate Exposure ... Small ... 32
Monitor ... Moderate Exposure ... Medium ... 13
Monitor ... Low Exposure ... Medium ... 4
Priority 2 ... Moderate Exposure ... Large ... 4
Priority 1 ... High Exposure ... Large ... 3
Priority 1 ... High Exposure ... Medium ... 1
Priority 2 ... High Exposure ... Small
Workforce Assessment and Communication
* Conduct skills inventories in high-exposure counties (composite scores above 60) to identify workers in vulnerable occupations.
* Launch public information campaigns explaining AI as a "change multiplier" rather than a job-loss multiplier, emphasizing augmentation over replacement.
* Partner with higher education institutions to assess current curriculum gaps in AI literacy.
Indiana University's GenAI 101 course, open to students, faculty, and staff, is one useful model; the Subcommittee and this bill's provisions could help identify how models like it might be extended beyond a single institution's community.
Targeted Support for High-Risk Workers
* Prioritize retraining and upskilling programs for occupations with high AI exposure.
* Focus on counties with both high AI exposure scores and large employment bases for maximum impact, without overlooking more rural areas.
Key insight: the moderate range of county AI-exposure scores, from roughly 48 at the low end to roughly 63 at the high end, suggests Indiana has time to adapt thoughtfully rather than react in crisis mode. This county-level variation offers a roadmap for targeted, regionally informed interventions rather than a one-size-fits-all approach.
* * *
Original text and figures here: https://www.help.senate.gov/imo/media/doc/6cbbd241-b14f-d7fb-a82f-91145ad11dc1/Rogers%20Testimony_9313ec8a-05aa-47be-a28b-058761307e2d.pdf
* * *
I. Introduction and Qualifications
Chairman Banks, Ranking Member Hickenlooper, and Members of the Subcommittee, thank you for the opportunity to submit this written statement in connection with the Subcommittee's hearing, "The Impact ... Show Full Article WASHINGTON, Aug. 19 -- The Senate Health, Education, Labor and Pensions Subcommittee on Employment and Workplace Safety released the following written testimony by Carol O. Rogers, director of the Indiana Business Research Center at the Indiana University Kelley School of Business, from a July 29, 2026, hearing entitled "The Impact of AI on the Workforce": * * * I. Introduction and Qualifications Chairman Banks, Ranking Member Hickenlooper, and Members of the Subcommittee, thank you for the opportunity to submit this written statement in connection with the Subcommittee's hearing, "The Impactof AI on the Workforce," and specifically on the AI Workforce PREPARE Act (S. 3339).[11]
I am Carol O. Rogers, Director of the Indiana Business Research Center (IBRC) at the Indiana University Kelley School of Business. I am here today as a subject matter expert; the views I express are my own and do not represent my employer, Indiana University.
As Director of IBRC, I work on issues related to Indiana's population, economy, and workforce to help Hoosier policymakers, educators, employers, and workers. In that role, my staff and I work daily with federal statistical series, including data from the Bureaus of Labor Statistics, Census, and Economic Analysis, as well as with Indiana's state administrative records, including unemployment insurance wage records and the state's longitudinal education-to-workforce data system. I have also served on the boards of several national research organizations, including as chairman of national organizations including the Labor Market Information Institute and the Council for Community and Economic Research.
I offer this testimony not as an advocate for a particular outcome on AI policy, but as a researcher with direct, practical experience in the strengths and limitations of the federal and state data infrastructure this bill would rely on. My comments focus on the available evidence showing, as best it can, AI's impact on the workforce, and on whether and where the underlying data infrastructure is ready to support this bill's goals.
II. What We Are Seeing: AI's Impact on the Workforce So Far
Generative and agentic AI has evolved quickly since ChatGPT's public release in late 2022.
Adoption has moved from early experimentation to routine business use: nearly 20 percent of U.S. firms now report using AI in at least some business function within the past two weeks (based on the latest data from February 2026), according to the U.S. Census Bureau's Business Trends and Outlook Survey.[1] Individual adoption has grown even faster. Harvard's Project on Workforce finds that 62 percent of U.S. adults ages 18 to 64 now use generative AI, and that 6.3 percent of total work hours are spent using it,[2] building on earlier academic research documenting the pace of generative AI's diffusion into daily work and jobs.[3]
* * *
Figure 1. Share of firms currently using AI in some business function, by state, February 2026.
Source: U.S. Census Bureau, Business Trends and Outlook Survey (BTOS), census.gov/hfp/btos.
* * *
This state-level detail from BTOS, shown in Figure 1, illustrates the unevenness of AI adoption across the country. Current AI use among firms ranges from a low of 11.1 percent in Arkansas to a high of 28.8 percent in Massachusetts, a gap of more than seventeen percentage points. Idaho (28.1 percent), Virginia (27.6 percent), Nevada (27.3 percent), and Washington (26.2 percent) round out the highest-adoption states, while Mississippi (11.8 percent), Nebraska (12.3 percent), New Mexico (13.6 percent), and Alabama (14.3 percent) report the lowest current use. A few patterns stand out. Adoption tends to run highest in states with large technology and professional-services sectors, such as Massachusetts, Virginia, and Washington, or with unusually concentrated industry clusters, as in Idaho and Nevada, and lowest in states with economies weighted toward agriculture and traditional manufacturing, concentrated in the Great Plains and Deep South. Neighboring states can also diverge sharply: Virginia's 27.6 percent sits beside West Virginia's 16.1 percent, and Idaho's 28.1 percent borders Oregon's 18.9 percent, a reminder that state averages can mask very different local economies just across a state line.
Indiana, at 20.9 percent, sits close to the roughly 20 percent national rate cited above, indicating that Indiana's experience is broadly representative of the national middle rather than an outlier in either direction.
My Center recently published research from the Indiana Department of Workforce (In Context, Mar-Apr 2026) comparing occupations with high and low AI exposure.[4] That analysis found job postings in AI-exposed occupations in Indiana shrank 41 percent since 2022, compared to a 35 percent decline in less-exposed occupations. Advertised wages in exposed occupations grew faster over the same period, 16 percent versus 11 percent, and the unemployment-rate advantage AI-exposed workers once held, a full percentage point in 2022, had nearly disappeared by last fall. Taken together, this looks less like mass job elimination and more like market-based repricing: some occupations losing ground, others gaining pay, often within the same broad category.
This pattern is consistent with what we are seeing nationally. The Federal Reserve Bank of Atlanta's survey of more than 700 corporate executives found little near-term aggregate employment decline attributable to AI, alongside a compositional reallocation of labor, with administrative support roles declining as demand grows for skilled technical roles.[5] The Federal Reserve Bank of Dallas has documented a similar wage pattern: nominal wages are up 7.5 percent across all industries since late 2022, but 8.5 percent in the most AI-exposed industries, and nearly 17 percent in computer systems design specifically.[6] Federal Reserve Governor Michael Barr summarized this dynamic in a February 2026 address, noting that AI's long-run effects on the labor market are likely to be positive, but that in the short term it can meaningfully disrupt specific workers, making how we manage the transition consequential.[7] At the same time, I would caution everyone against inferring too much from Amazon, Meta, and other major technology firms that have announced substantial AI-linked workforce reductions over the past year. Outplacement firm Challenger, Gray & Christmas reports that AI has been cited in roughly a fifth to a quarter of all announced U.S. layoffs so far in 2026.[8] But that pattern is concentrated in a narrow slice of large, capital-intensive technology firms making enormous, simultaneous investments in AI infrastructure. It is not what we see broadly across small and medium-sized businesses. The National Federation of Independent Business's most recent survey finds that 98 percent of small employers currently using AI report no change in their employee counts, and NFIB's monthly economic trends data show small business hiring plans holding at a modest net positive.[9] Policy built primarily around the experience of the largest companies risks missing what is actually happening for the overwhelming majority of American employers.
Finally, a broader macro signal deserves attention. Goods-producing employment declined over the course of last year and has largely plateaued in 2026, even as productivity and wages have continued to rise.[10] Fewer workers producing more value, at higher pay, is consistent with a meaningful contribution from automation and AI, working alongside other factors such as capital investment, though this causal link is not yet provable with the data currently available. It is exactly the kind of signal this bill's data infrastructure should help us track and understand with more precision than we can today.
III. Why Data Infrastructure Will Determine This Bill's Success
The AI Workforce PREPARE Act is, at its core, a data bill. Its central premise, stated plainly in its findings, is that policymakers, training providers, and workers currently lack the data and forecasts needed to anticipate and mitigate AI-driven worker dislocation.[11] I agree with that premise, and the evidence summarized in Section II above illustrates why: the picture is real, but it is also mixed, still emerging, and in places contradictory. Where I would like to focus the Subcommittee's attention is a second-order question the bill does not fully resolve: whether the federal and state data systems it directs agencies to use and improve are themselves currently capable of producing timely, reliable, AI-attributable signals about the labor market.
In my experience, the answer is mixed. Some of the infrastructure this bill would lean on (particularly state administrative wage records), is more capable than is generally understood in federal policy discussions, but underutilized. Other pieces, particularly the timeliness of occupational classification systems and the standardization of employer self-reported data, are not yet fit for the purpose this bill assigns them. Both of these points matter for implementation, funding, and oversight.
IV. Section-by-Section Assessment
A. AI Workforce Research Hub
The bill's creation of an AI Workforce Research Hub to convene researchers, employers, and labor, and to help implement the White House AI Action Plan, is a sound organizing structure.
Its value will depend heavily on whether it is designed to integrate existing state and federal data assets rather than to build a new, parallel data collection effort from scratch. States like Indiana have already made substantial investments in longitudinal data infrastructure; a federal hub that does not connect to this work risks duplicating effort and producing a less complete picture than what already exists in pieces across federal agencies and states.
B. WARN Act Amendment -- The "Substantial Factor" Standard
Amending the WARN Act to require disclosure when AI is a "substantial factor" in a mass layoff is a meaningful transparency measure, and I understand its intent. However, as a federal data consumer, I want to flag a specific risk: the bill directs the Secretary of Labor to issue guidance on how employers may determine that AI is a substantial factor and estimate the associated share of job loss, but employer compliance is explicitly satisfied by a "good-faith statement," without a standardized measurement method attached to it.[11]
Absent a common framework, I would expect significant variation across employers and industries in how "substantial factor" is interpreted, not necessarily out of bad faith, but because there is currently no agreed-upon, operational definition connecting a specific business decision to a specific automation event. This matters because data that cannot be standardized across employers cannot be aggregated into a reliable and comparable national and regional picture, which undercuts the purpose of collecting it and could lead to thousands of differing interpretations.
Recommendation: DOL's guidance should point employers toward existing, standardized classification systems, such as the Standard Occupational Classification (SOC) system and O*NET tasks and skills data, as a reference framework for these determinations, so that reported information is at least structurally comparable across firms, sectors, and states, and can be crosschecked against independent administrative data and other sources over time rather than resting solely on self-reporting.
C. BLS Occupational Forecasts for AI-Sensitive Occupations
BLS occupational projections, and the Standard Occupational Classification system that underpins them, update on multi-year cycles that are considerably slower than the pace at which AI is changing specific tasks and skills within occupations. An occupation can appear stable in the aggregate data long after AI has meaningfully altered the tasks performed within it and the skills required by the worker. I would encourage the Subcommittee to direct BLS toward task-level analysis, for example building on the O*NET task and skills content with more frequent employer surveys, rather than relying solely on occupation-level projections, which are a blunter instrument for this specific policy question.
In terms of resources, in my experience with comparable state and federal data modernization efforts, the anticipated funding over a five-year period is unlikely to be sufficient to meaningfully accelerate BLS's forecasting capacity for such a fast-moving technology. I would recommend the Subcommittee revisit this authorization level as implementation proceeds and consider it a floor rather than an adequate ceiling.
D. Researcher Access to Federal Workforce Data
The bill's provision to increase researchers' access to federal workforce data is, in my assessment, one of its most consequential and most under-discussed elements. Speaking from direct experience navigating Federal Statistical Research Data Center (FSRDC) access, restricted-use microdata agreements, and interagency data-sharing arrangements, I can attest that current approval timelines, inconsistent disclosure review standards across agencies, and limited capacity for linking federal data with state administrative records are all real, practical barriers to timely research, not merely inconveniences.
I would encourage the Subcommittee to seek specificity in how this provision is implemented: whether it means expanded FSRDC access, streamlined restricted-use data agreements, funding for state-federal data linkage infrastructure, or some combination. Without that specificity, this provision risks remaining a statement of intent rather than a change we can actually use.
E. State Administrative Records and In-Demand Occupation Lists
This is the area where I can speak most directly from operational experience, and where I believe the Subcommittee has the greatest opportunity to strengthen the bill.
States' unemployment insurance wage records, together with Statewide Longitudinal Data Systems (SLDS) that link education and workforce outcomes, already capture actual, verified employment and earnings changes on a quarterly basis, at a level of industry and geographic granularity that federal survey instruments cannot match. These records are, in effect, one of the most complete real-time pictures of the American labor market already in existence, and they are collected and maintained by every state, largely independent of federal AI-specific policy. In Indiana, we use these records daily to inform workforce and education policy, and I believe they may be underappreciated in Washington relative to their value.
Two considerations follow. First, I would recommend the bill explicitly direct the AI Workforce Research Hub and related data-collection provisions to connect with, rather than duplicate, this existing state infrastructure, including funding mechanisms for states to modernize and link these records, building on ongoing Enhanced Wage Records initiatives, for AI-workforce intelligence purposes. Second, I want to be candid that data definitions, industry coding practices, and system capacity vary meaningfully from state to state. This is a real and, I believe, solvable technical barrier, but it will require dedicated federal investment and technical assistance to states, not simply a data-sharing mandate, to produce comparable results across states, regions, and even congressional districts in ways that truly inform policy.
Finally, I want to flag the stakes of getting this right: the bill ties AI-informed labor market projections to states' in-demand occupation lists, which in turn determine which training programs receive federal workforce development funding under WIOA and related authorities.
This creates a direct chain from federal AI-workforce data quality to aligning which training program an individual worker is funded to attend. Forecast error at the federal level does not stay abstract; it becomes a misallocated training dollar and, for an individual worker, a missed opportunity.
F. Voluntary Public-Private Data Sharing and Prize Competitions
The bill's approach to facilitating voluntary data sharing from AI developers and running prize competitions to better understand AI adoption and task-level automation is a reasonable, low-friction starting point, particularly relative to more prescriptive mandatory-disclosure proposals also pending before Congress. My caution here is methodological: voluntarily submitted adoption data, whether from employers or AI companies, is likely to be subject to selection bias, both in which firms choose to participate and in how they characterize their own AI use. I would recommend that any resulting datasets be explicitly validated against independent administrative benchmarks, such as those described in Section IV.E, rather than treated as authoritative on their own.
V. My Experience in Indiana
My work at IBRC sits directly at the intersection of federal statistics and Indiana's own administrative records, and demand across the state for actionable intelligence on AI's impact is high. At the same time, standardized, complete, current, local data remains hard to come by.
Businesses ask us for competitive intelligence. Workers want career and training guidance in an AI-uncertain world. Policymakers want to know whether Indiana will have the workforce employers need. Educators need to align curricula.
We are estimating impact now using AI-exposure metrics by Indiana county, but we still do not know how employers are actually implementing AI, governing its use, integrating it into individual job tasks, or what specific AI skills they are hiring for, training toward, or planning around, the detail we need to gauge impact accurately. This is precisely the kind of employer-level, task-level information the researcher-access and data-hub provisions of this bill could help supply, if implemented with the state-data linkages described in Section IV.E above.
I do not come before this Subcommittee as an advocate for a predetermined conclusion about AI's impact, but as someone whose job it is to make sure Indiana's workers, educators, and policymakers know what is happening, and what is likely to happen, to our workforce.
VI. Recommendations
In summary, I offer the following recommendations for the Subcommittee's consideration as it advances this bill:
* * *
1. Direct BLS to prioritize task- and skill-level analysis, building on O*NET, alongside occupation-level projections, given the pace at which AI is changing work within occupations faster than occupational classifications themselves change.
* * *
2. Require DOL's WARN Act guidance to anchor employer "substantial factor" determinations to standardized occupational, skill, and task classifications, so reported data can be aggregated and compared across employers and states.
3. Revisit the funding authorization for BLS occupational forecasting as a floor, not a sufficient level of investment, given the scope of the task.
4. Specify what expanded researcher access to federal workforce data will concretely mean in practice, whether FSRDC capacity, restricted-use data agreement timelines, or state-federal linkage funding, rather than leaving the provision as a general statement of intent.
5. Explicitly direct the AI Workforce Research Hub and related provisions to connect with existing state administrative data infrastructure, including UI wage records as part of ongoing Enhanced Wage Records initiatives and Statewide Longitudinal Data Systems, and fund state-level technical assistance to improve cross-state comparability, rather than building new, parallel federal data collection.
6. Require independent validation of voluntarily submitted AI-adoption data against administrative benchmarks before it informs funding-linked determinations such as states' in-demand occupation lists.
7. Direct the AI Workforce Research Hub to develop a standardized instrument for capturing employer-level AI implementation, governance, and skills-demand data, since neither federal nor state administrative systems currently capture this information at the granularity needed to guide training investment.
VII. Conclusion
The bill reflects a sound, bipartisan recognition that this country needs better data to manage the workforce transitions AI is already producing. My purpose today has been to bring a research and data-infrastructure perspective to that effort: to describe what the available evidence currently shows, to highlight where existing federal and state data systems are ready to meet this bill's ambitions, and to flag where targeted investment and design choices will determine whether they can. I would welcome the opportunity to work with the Subcommittee and its staff as this legislation moves forward, and I am happy to answer any questions.
* * *
References
[1] U.S. Census Bureau. Business Trends and Outlook Survey (BTOS). 2026. census.gov/hfp/btos.
[2] Bick, Alexander, and Adam Blandin. GenAI Adoption Tracker. Data as of May 2026. genaiadoptiontracker.com.
[3] Bick, Alexander, Adam Blandin, and David J. Deming. "The Rapid Adoption of Generative AI." NBER Working Paper No. 32966. National Bureau of Economic Research, 2024 (revised 2026).
[4] InContext, "Is AI Affecting Indiana's Labor Market?" March-April 2026. Indiana Business Research Center, incontext.indiana.edu.
[5] Federal Reserve Bank of Atlanta. "How Might AI Change the Workplace? Evidence from Corporate Executives." Policy Hub, March 25, 2026; see also Working Paper 202604, "Artificial Intelligence, Productivity, and the Workforce: Evidence from Corporate Executives." atlantafed.org.
[6] Davis, J. Scott. "AI Is Simultaneously Aiding and Replacing Workers, Wage Data Suggest." Federal Reserve Bank of Dallas, February 24, 2026. dallasfed.org/research/economics/2026/0224.
[7] Barr, Michael S. "What Will Artificial Intelligence Mean for the Labor Market and the Economy?" Speech at the New York Association for Business Economics, February 17, 2026. federalreserve.gov.
[8] Challenger, Gray & Christmas, Inc. Monthly Job Cuts Report, 2026 (AI cited in approximately 22-23 percent of announced U.S. layoffs year-to-date as of mid-2026). challengergray.com.
[9] National Federation of Independent Business. "Small Business and Technology Survey." June 25, 2025; and NFIB Small Business Economic Trends (SBET) monthly reports, 2026. nfib.com.
[10]U.S. Bureau of Labor Statistics. Employment Situation and goods-producing employment series (CES), 2025-2026, retrieved via FRED, Federal Reserve Bank of St. Louis. fred.stlouisfed.org.
[11] AI Workforce PREPARE Act, S. 3339, 119th Congress (2025-2026). congress.gov/bill/119th-congress/senate-bill/3339.
* * *
Supplementary material appears below that is part of an Indiana Business Research Center working paper looking at AI exposure throughout Indiana, by county. I provide this as further insight into the level of detail sought in my state to understand the breadth of AI potential for employers and workers, as well as noting the potential risks.
Generative AI Exposure Assessment: Indiana Counties
An IBRC Workforce and Economic Development Analysis Working Paper
Indiana Business Research Center, Indiana University Kelley School of Business
* * *
Figure. Intensity of AI-exposed jobs across Indiana counties.
Source: IN Context, Mar-Apr 2026, Indiana Business Research Center, incontext.indiana.edu.
* * *
The figure above shows the county-level detail behind an estimate of the exposure to Indiana's occupations. Exposure scores cluster tightly across Indiana's 92 counties, ranging from roughly 52 at the low end to 60 at the high end, an eight-point spread, with a median near 56. Most counties fall within a narrow band close to that median, while a handful of counties scattered across both the northern and southern parts of the state register the highest scores on the map, and a single county in the north-central part of the state registers the lowest. This tight, dispersed clustering suggests AI exposure in Indiana is not confined to the state's largest metropolitan counties, but runs at meaningfully similar intensity across metropolitan, small-city, and rural counties alike, underscoring why sub-state, county-level data, and not just statewide or occupational averages, is necessary to guide training investment and workforce policy.
Methodology
Data Integration:
* Lightcast(TM) Employment Data: 2.7 million Indiana jobs across 909 industries.
* County Staffing Patterns: Occupation-industry matrices for each county.
* Felten AI Scores: Generative AI exposure by occupation and industry.
Key Innovation: Rather than applying generic industry scores, this analysis accounts for countyspecific occupational compositions. The same industry can have different AI exposure across counties based on local staffing patterns.
Two types of AI scores are combined to produce a composite score for each county and industry:
* Image Generation AI Exposure: measures how susceptible occupations are to being augmented or replaced by AI systems that create visual content, such as DALL-E, Midjourney, or Stable Diffusion. High scores indicate jobs where workers currently spend significant time on tasks like graphic design, illustration, photo editing, or visual content creation that AI image generators can now perform.
* Language Modeling AI Exposure: measures exposure to AI systems that understand and generate text, such as ChatGPT, Claude, or GPT-5. High scores indicate occupations involving substantial writing, analysis, research, communication, or information-processing tasks that large language models can augment or automate.
The two scores are combined into a composite AI score, an employment-weighted average of the two types of scores. The full calculation flow:
1. Occupation level: each occupation has an image-generation score and a language-modeling score, on a 0-100 scale.
2. Industry level: for each county-industry combination, each occupation's AI scores are weighted by its employment share within that industry, and the two AI types are averaged.
3. County level: each industry's composite score is weighted by its employment share within the county, producing a final, employment-weighted county score.
These scores measure AI risk: these scores focus specifically on generative AI capabilities that have rapidly advanced over the past two to three years, rather than on traditional automation or robotics. The underlying Felten et al. research identifies which job tasks are most exposed to these newer AI capabilities, providing a more current and relevant assessment of workforce disruption potential than older automation studies. Complementary coverage: image generation primarily affects creative, design, and visual-communication roles, while language modeling affects knowledge work, analysis, and text-based tasks. Together, they capture the breadth of generative AI's current capabilities and provide a fuller view of which occupations face the most immediate AI-driven change.
Coverage: 92 Indiana counties, representing 2,790,788 jobs.
Statewide Patterns
There is a strong positive relationship between counties with high language-modeling exposure and high image-generation exposure.
* Most of the low-exposure counties are rural, with the exception of Boone County, which is located in the Indianapolis metropolitan statistical area.
* There are fewer clear geographic patterns among the high-exposure counties, though most are part of metropolitan areas.
* Martin County stands out as the county with the highest AI exposure score in the state.
This small county's high score is driven by its outsized employment share, approximately 29 percent, in Engineering Services (NAICS 541330), a location quotient of 8.5, an industry with a high image-generation exposure score based on its local staffing patterns.
* Ohio County presents a useful counterexample. Given its high employment in hospitality, driven by the Rising Sun Casino Resort, it was surprising that this county also scored highly on AI exposure. This underlines an important caution for the Subcommittee: these exposure scores should not be relied on blindly for workforce planning. Local knowledge remains essential to interpreting them correctly.
Combining exposure level with county employment size points to a practical, tiered approach for workforce-development priority:
Priority Level ... Exposure Tier ... Employment Size ... Counties
Monitor ... Low Exposure ... Small ... 34
Monitor ... Moderate Exposure ... Small ... 32
Monitor ... Moderate Exposure ... Medium ... 13
Monitor ... Low Exposure ... Medium ... 4
Priority 2 ... Moderate Exposure ... Large ... 4
Priority 1 ... High Exposure ... Large ... 3
Priority 1 ... High Exposure ... Medium ... 1
Priority 2 ... High Exposure ... Small
Workforce Assessment and Communication
* Conduct skills inventories in high-exposure counties (composite scores above 60) to identify workers in vulnerable occupations.
* Launch public information campaigns explaining AI as a "change multiplier" rather than a job-loss multiplier, emphasizing augmentation over replacement.
* Partner with higher education institutions to assess current curriculum gaps in AI literacy.
Indiana University's GenAI 101 course, open to students, faculty, and staff, is one useful model; the Subcommittee and this bill's provisions could help identify how models like it might be extended beyond a single institution's community.
Targeted Support for High-Risk Workers
* Prioritize retraining and upskilling programs for occupations with high AI exposure.
* Focus on counties with both high AI exposure scores and large employment bases for maximum impact, without overlooking more rural areas.
Key insight: the moderate range of county AI-exposure scores, from roughly 48 at the low end to roughly 63 at the high end, suggests Indiana has time to adapt thoughtfully rather than react in crisis mode. This county-level variation offers a roadmap for targeted, regionally informed interventions rather than a one-size-fits-all approach.
* * *
Original text and figures here: https://www.help.senate.gov/imo/media/doc/6cbbd241-b14f-d7fb-a82f-91145ad11dc1/Rogers%20Testimony_9313ec8a-05aa-47be-a28b-058761307e2d.pdf
American Bankers Association Executive VP Benda Testifies Before Senate Special Committee on Aging
WASHINGTON, Aug. 19 -- The Senate Special Committee on Aging released the following testimony by Paul Benda, executive vice president for risk, fraud and cybersecurity at the American Bankers Association, from a July 29, 2026, hearing entitled "The AI Deception Machine: Deepfakes, Chatbots, and the New Frontier of Senior Fraud":
* * *
Chairman Scott, Ranking Member Gillibrand, and members of the Committee, thank you for the opportunity to testify for today's hearing entitled: "The AI Deception Machine: Deepfakes, Chatbots, and the New Frontier of Senior Fraud." My name is Paul Benda, and I serve ... Show Full Article WASHINGTON, Aug. 19 -- The Senate Special Committee on Aging released the following testimony by Paul Benda, executive vice president for risk, fraud and cybersecurity at the American Bankers Association, from a July 29, 2026, hearing entitled "The AI Deception Machine: Deepfakes, Chatbots, and the New Frontier of Senior Fraud": * * * Chairman Scott, Ranking Member Gillibrand, and members of the Committee, thank you for the opportunity to testify for today's hearing entitled: "The AI Deception Machine: Deepfakes, Chatbots, and the New Frontier of Senior Fraud." My name is Paul Benda, and I serveas Executive Vice President for Risk, Fraud and Cybersecurity at the American Bankers Association. ABA is the voice of the nation's banking industry, which is composed of small, regional and large banks that together employ more than two million people, safeguard trillions of dollars in deposits, and extend credit in communities across the country.
Banks play a unique and essential role in supporting the financial health of older Americans. As trusted institutions deeply rooted in their communities, banks are often the first, and most consistent, financial touchpoint for seniors and their families. Every day, bank employees help older customers manage retirement income, protect their savings, recognize and avoid scams, and adapt to an increasingly digital financial system. These relationships position banks not only as providers of financial services, but also as front line educators and protectors of financial security.
Banks have long used advanced analytics and artificial intelligence to identify suspicious transactions, protect customer accounts, and respond to cyber threats. Generative AI is strengthening those defensive tools. At the same time, criminals are using the same technology to make familiar scams more convincing, more personalized, cheaper to launch, and easier to scale.
The central point of my testimony is straightforward: generative AI is not replacing traditional scams. It is industrializing them. A criminal can now create a convincing voice, video, photograph, text message, advertisement, or online persona with little technical skill and at very low cost. AI allows fraudsters to imitate a friend, family member, bank employee, government official, physician, pastor, financial adviser, or virtually anyone else whose identity may inspire trust.
Older Americans face especially serious consequences from these scams. A loss may represent retirement savings that cannot be replaced, proceeds from a home sale, or money set aside for medical and long-term care needs. The financial damage can be compounded by shame, isolation, fear of losing independence, and reluctance to report the crime. We should be careful not to describe older Americans as inherently unsophisticated. These crimes succeed because professional criminals deliberately exploit trust, fear, urgency, family relationships, and authority. Generative AI makes those tactics substantially more credible.
My testimony focuses on four issues. First, I will describe how generative AI is changing the nature of scams. Second, I will explain why those changes are particularly significant for older Americans. Third, I will discuss what banks and ABA are doing to protect customers. Finally, I will offer practical steps for consumers and targeted policy recommendations for Congress and federal agencies.
Fraud Has Become an International Criminal Industrial Complex Today's scams are often not isolated crimes committed by a lone opportunist. They are carried out by sophisticated domestic and transnational criminal networks that use stolen personal information, fake identities, money mules, fraudulent advertisements, spoofed calls and texts, fake websites, encrypted communications, and rapid payment channels. The infrastructure that supports these crimes already exists, and generative AI is now being layered onto that infrastructure, making it even more effective.
AI can help criminals identify likely victims, collect personal information from public sources and data breaches, create customized scripts, translate communications into multiple languages, and sustain conversations with many victims at once. It can generate polished messages without the grammatical errors or awkward phrasing that once served as warning signs of a scam. It can also create synthetic identity documents and automate efforts to open accounts or take over existing accounts.
A recently published joint report by the ABA, Better Identity Coalition, and the Financial Services Sector Coordinating Council (FSSCC), Mitigating AI-Powered Attacks Against Identity and Authentication,/1 identifies three broad AI-enabled attack vectors: deepfake-driven social engineering and impersonation, synthetic identity creation, and the use of AI agents as attack surrogates capable of automating account takeover and fraud activity. These threats affect both consumers and financial institutions, but for many older Americans the most immediate danger is the use of AI to make social engineering and impersonation scams more convincing, personalized, and difficult to detect.
How Generative AI Is Changing Scams Against Older Americans
The fundamental danger is that seeing or hearing someone can no longer be treated as proof of identity. A voice message may sound like a loved one. A video may appear to show a government official. A caller ID display may show the correct name and phone number of a bank. A social media account may use a familiar photograph and a convincing history of posts.
Any of these signs of authenticity can be fabricated or manipulated.
Consumers have traditionally been advised to look for poor grammar, unnatural speech, obvious image defects, or inconsistent stories. Those remain useful clues, but they are becoming less reliable. A scam does not need to be technically perfect. It only needs to be credible enough, particularly when combined with urgency, secrecy, stolen personal information, and emotional manipulation.
A criminal can use information from social media, public records, obituaries, data brokers, breached accounts, and prior scam reports to tailor an approach. The scammer may know the names of family members, details from a recent trip, a medical condition, the victim's bank, or the identity of a trusted professional. That information can be woven into an AI-generated message or conversation, turning a generic scam into a seemingly personal emergency.
Generative AI enables a single criminal organization to maintain many apparent relationships at the same time. It can produce messages around the clock, adapt responses based on what a victim says, and translate content rapidly. This capability is especially dangerous in romance, investment, and recovery scams, where criminals may spend weeks or months building trust with their targets before requesting money.
A single fraud may begin with a social media advertisement, continue through text messages, move to a phone or video call, direct the victim to a fake website, and end with a bank transfer, cryptocurrency purchase, cash withdrawal, or courier pickup. Each channel can appear to confirm the others even when the entire experience is controlled by the same criminal group.
* * *
1 Financial Services Sector Coordinating Council, Mitigating AI-Powered Attacks Against Identity and Authentication (February 2026), https://fsscc.org/wp-content/uploads/2026/02/AI-IA-Workstream-Mitigations.pdf
* * *
A Recent FBI Warning Shows the Threat in Practice
A July 20, 2026, public service announcement from the FBI's Internet Crime Complaint Center provides a timely example of how generative AI is being used to enhance and scale fraud schemes. The FBI warned that criminals are impersonating FBI and IC3 personnel to revictimize people who have already lost money. These scams use fake social media profiles, spoofed government websites, and AI-generated videos that depict senior FBI personnel encouraging victims to submit information through a fraudulent IC3 website./2
The alert is particularly important because it incorporates many of the defining features of this emerging threat. Criminals target prior victims, impersonate trusted authorities, use AI-generated videos to create a false sense of legitimacy, distribute content through social media, and direct victims to websites designed to collect personal and financial information. The promise of recovering stolen funds is itself a scam designed to steal even more money.
The FBI also warns that scammers may use AI-generated video during real-time chats to impersonate executives, law enforcement officials, or other authority figures. This illustrates why consumers cannot rely solely on the apparent realism of a video or voice. The correct response is independent verification through a trusted channel.
Scam Scenarios Particularly Relevant to Older Americans
Grandchild and family-emergency scams - A victim receives a call that sounds like a child, grandchild, or other relative in distress. The caller claims to have been arrested, injured, kidnapped, or hospitalized. A second person may pose as a lawyer, police officer, doctor, or court official. The victim is instructed to send money immediately and not contact other family members. Voice cloning can remove one of the most important reasons a victim might otherwise question the call.
Government and law enforcement impersonation - A criminal may claim to represent the FBI, Social Security Administration, IRS, Medicare, a local sheriff, or another agency. The victim may be told that benefits will be suspended, that an arrest is imminent, that an identity has been compromised, or that money must be transferred for safekeeping. The same tactic may be used in recovery scams, where the criminal falsely promises to recover an earlier loss in exchange for a fee or additional payment.
Bank-impersonation scams - A victim receives a text or call that appears to come from the bank's fraud department. The caller may know partial account information, refer to a supposed transaction, or cause the bank's actual number to appear on caller ID. The criminal may ask the victim to disclose an authentication code, approve a login, install remote-access software, or move funds to a supposed "safe account." The FCC reinforced this warning in a July 21, 2026, consumer alert,/3 noting that bank-impersonation scams account for the highest losses of any impersonation scam category and that victims may lose everything in their accounts. The FCC emphasized a simple rule: banks will never call or text consumers and ask them to move money to protect it.
* * *
2 Federal Bureau of Investigation, Internet Crime Complaint Center, "FBI Warns of Scammers Impersonating the IC3," Alert No. I- 072026-PSA (July 20, 2026), https://www.ic3.gov/PSA/2026/PSA260720.
3 Federal Communications Commission, FCC Consumer Alert: Bank Impersonation Scams (July 21, 2026).
* * *
A recent bank survey illustrates the scale of the problem. Among 14 surveyed large banks, the average number of identified bank-impersonation scams increased 150 percent from 2024 to 2025, reaching an average of 26,196 scams per bank. Telecom-originated scams increased 124 percent, while scams originating on social media more than doubled.
The survey is not a census of the entire banking industry, but it provides strong directional evidence from institutions representing a significant share of industry assets./4
The same survey found that social media companies took nearly two weeks on average to remove bank-impersonation content and did not respond to more than 20 percent of takedown requests. Two surveyed banks reported that 40 percent of calls appearing to originate from bank-owned numbers had been spoofed. These results underscore that banks cannot stop impersonation scams alone. The deception often reaches the consumer through a telecom network, social media platform, advertising system, or messaging service before a bank sees any transaction./5
Romance and investment scams - AI-generated photographs, voice messages, video chats, and chatbots can help criminals build and sustain false relationships over long periods of time. A single criminal organization may manage many such relationships simultaneously. The relationship may eventually be used to persuade a victim to invest in a fraudulent scheme, send money in response to a manufactured emergency, receive and transfer funds, or participate in another form of financial exploitation. The emotional bond can make intervention extraordinarily difficult, even when family members or bank employees identify clear warning signs.
Technical-support and account-security scams - A criminal may impersonate a technology company, bank security specialist, or government cybersecurity official and claim that the victim's device or account has been compromised. The victim may be directed to install remote-access software, share a screen, reveal an authentication code, or transfer funds. AI-generated websites, instructions, and voices can make the approach look professional and legitimate.
Recovery and repeat-victimization scams - Prior victims are often targeted again by people claiming to be lawyers, investigators, banks, cybersecurity firms, or recovery services. The victim is promised that stolen funds have been located but must first pay a fee, tax, or processing charge. The recent FBI alert specifically warns about this form of "double-dip" scam. It is especially harmful because a prior victim may be emotionally vulnerable and desperate to recover retirement savings.
Why the Consequences for Older Americans Can Be Especially Severe The financial impact of a scam can be devastating at any age, but older victims may have less time and fewer employment opportunities to rebuild savings. Losses may involve retirement assets, home equity, emergency funds, or money needed for health and long-term care.
* * *
4 Tara Payne, Drew Ruben & Madison Stulga, Bank Policy Institute, "Hijacking Trust: How Scammers Exploit the Banking System's Most Valuable Asset" (July 20, 2026), https://bpi.com/impersonation/. The survey covered 14 BPI member banks ranging from approximately $100 billion to more than $1 trillion in assets.
5 Id.
* * *
Victims may also hesitate to report. Some fear embarrassment or worry that family members will question their ability to manage their own finances. Others remain under the criminal's influence and believe that one more payment will resolve the emergency or recover prior losses. In romance scams, the victim may remain emotionally attached to the person they believe is real.
These dynamics make scam intervention different from ordinary transaction monitoring. Bank employees sometimes encounter customers who have been coached to lie about the purpose of a transaction, conceal communications, or distrust the very bank employee trying to protect them.
Effective intervention therefore requires empathy, training, and an understanding of the psychological manipulation used by professional criminals.
How Banks Are Protecting Customers
Banks employ a host of tools to detect and deter fraud, including AI, machine learning, and other advanced analytics to identify unusual transactions, new or suspicious payees, account takeover, abnormal device activity, behavioral anomalies, manipulated documents, and high-risk payment patterns. These systems can surface transactions for additional review or prompt the bank to contact the customer before a transaction is completed.
While criminals use AI to make deception more effective, banks employ AI to detect anomalies and protect customers. The goal should not be to restrict beneficial defensive uses. It should be to strengthen safeguards, improve information sharing, and reduce criminals' ability to exploit communications and identity systems.
Banks increasingly employ layered defenses that combine identity validation, device and behavioral signals, document analysis, liveness and injection-attack detection, out-of-band verification, additional confirmation for high-risk activity, and human review. There is no single "deepfake detector" that can solve the problem. Effective defense depends on multiple independent signals and risk-based controls.
Technology alone, however, cannot identify every scam. Banks also rely on human judgment because no model, algorithm, or automated system can reliably detect every form of deception.
Front-line bank employees are trained to recognize sudden payments to new recipients, unusual withdrawals, customers acting under extreme urgency, demands for secrecy, customers being coached over the telephone, and requests involving "safe accounts," cryptocurrency, gift cards, cash couriers, or other unusual payment methods.
ABA is expanding training that helps bankers recognize deepfakes and AI-enabled threats, support victims of financial crime, and intervene when a customer appears to be under a scammer's influence. That includes ABA's Spot the Scammer training and work focused on "breaking the scam spell," which recognizes that the challenge is not merely identifying an unusual transaction but helping a victim step outside the criminal's narrative./6
Fraud alerts and direct bank-customer communications are among the most effective tools available. A February 2026 Morning Consult poll commissioned by ABA found that nearly nine in ten bank customers said their bank takes proactive steps to protect them from fraud and scams.
* * *
6 American Bankers Association, "Spot the Scammer: How to Detect Scams, Deepfakes, and AI Threats," https://www.aba.com/training-events/online-training/spot-the-scammer.
* * *
Three in five consumers had received a fraud alert regarding potentially suspicious account activity, and 96 percent of those recipients found the alerts valuable./7
The same poll found that 72 percent of consumers believe banks do more than other industries to protect them from fraud and scams. By a margin of nearly five to one, consumers trusted banks more than other industries and the government to protect them from fraud. These findings are important because proposed legislation and the regulatory environment should support, rather than inadvertently impede, timely fraud alerts and other protective communications./8
Consumer education is an equally important component of fraud prevention. The ABA Foundation's Safe Banking for Seniors program provides banks with free, turnkey materials to educate older Americans, caregivers, and families about scams, identity theft, financial caregiving, and financial exploitation. More than 2,550 banks have participated in the program since it was launched in 2016.
ABA and the FBI have also developed consumer materials specifically addressing deepfake scams, and ABA maintains guidance explaining how deepfake media can be used in impersonation, investment, romance, employment, and other fraud schemes. ABA's #BanksNeverAskThat campaign teaches consumers to recognize common impersonation tactics, while the current "Snap Out of It!" theme focuses on interrupting the urgency and psychological manipulation used by scammers./9
ABA and the banking industry are working to improve the speed of bank-to-bank communications, standardize documentation used to request holds or returns of fraudulent funds, expand the ABA Fraud Contact Directory, and develop mechanisms for sharing fraud indicators.
ABA is also working with law enforcement and other partners to improve recovery of funds, including funds moved through cryptocurrency.
These efforts are critical because modern fraud schemes span multiple institutions and industries.
One bank may see the victim's outgoing payment, another may hold the receiving account, a telecom provider may carry the spoofed call, and a platform may host the fraudulent advertisement. No participant has the complete picture. Faster, privacy-protective sharing of fraud signals can allow earlier intervention and improve the chance of recovery.
What Older Americans and Their Families Can Do Consumer guidance should be memorable, focused on the mechanics of impersonation, and avoid requiring people to become deepfake experts. The goal is to change the decision process so that a convincing voice or video cannot by itself trigger a payment. ABA recommends that seniors and their families take the following steps to guard against fraud: Pause - Do not act immediately, no matter how urgent or frightening the communication appears. Time is one of the most effective defenses against an AI-enabled scam because pausing interrupts the criminal's script and creates an opportunity to verify the story.
* * *
7 Morning Consult National Tracking Poll commissioned by the American Bankers Association, conducted Feb. 21-25, 2026, among 4,456 U.S. adults; margin of error +/- 1 percentage point.
8 Id.
9 American Bankers Association, "Deepfake Media Scams," https://www.aba.com/news-research/analysis-guides/deepfake-media- scams; American Bankers Association, "ABA Foundation and FBI Release New Infographic to Help Americans Spot and Avoid Deepfake Scams," https://www.aba.com/about-us/press-room/press-releases/ABA-Foundation-and-FBI-Joint-Infographic-on- Deepfake-Scams.
* * *
Verify through a separate channel - Hang up and contact the person or organization using a telephone number or method already known to be legitimate. Do not use a callback number, link, or website supplied by the caller or message. Call another family member before sending money. Contact the bank using the number printed on the card or statement.
Use a family password - Families should agree in advance on a private word or phrase that can be used to verify a genuine emergency. The phrase should not be posted online or easily guessed. A family can also establish a personal question whose answer is not available through social media or public records.
Remember: "If it's a secret, it's a scam" - A demand to act now, stay on the line, or keep the matter secret is a powerful warning sign. Criminals use secrecy to prevent victims from contacting family, banks, or law enforcement. A legitimate bank, government agency, police department, or family member should not object to independent verification.
Do not trust the channel itself - Caller ID can be spoofed. Voices can be cloned.
Photographs and videos can be fabricated. Social media accounts can be copied or compromised. Seeing a familiar face or hearing a familiar voice is no longer sufficient proof of identity.
Never move money to a "safe account" - A bank or government agency will not direct a consumer to transfer funds to another account for protection. Requests involving cryptocurrency, gift cards, cash couriers, or payments to newly created accounts deserve particular scrutiny.
Involve a trusted person and report quickly - Before making a large or unusual payment, speak with a trusted family member, friend, banker, caregiver, attorney, or financial adviser. If money has already been sent, contact the bank immediately. Rapid reporting may improve the chance of stopping or recovering funds. Victims should also report the incident to the FBI's Internet Crime Complaint Center. The Senate Special Committee on Aging operates a Fraud Hotline that can help older Americans and families understand how to report scams and protect themselves./10
Most importantly, victims should not be shamed. These schemes are designed by professional criminals using sophisticated technology and psychological manipulation. Reporting quickly helps the victim and may help prevent others from being harmed.
Banks and Consumers Cannot Solve This Alone By the time a bank sees a transaction, a criminal may have communicated with the victim for days or months. The deception may have been enabled by a spoofed call, fraudulent text, paid advertisement, social media account, messaging application, fake website, or cryptocurrency platform. Responsibility must extend to the sectors that enable criminals to reach and manipulate victims.
The Morning Consult poll referenced above found that three in four Americans had seen fraudulent or deceptive advertisements on social media. Seventy-eight percent supported federal legislation requiring social media companies to do more to identify and remove fake accounts and fraudulent advertisements, and 77 percent supported regulatory action requiring telecommunications providers to do more to prevent spoofed caller IDs. Consumers clearly do not view scam prevention as solely a bank responsibility./11
* * *
10 U.S. Senate Special Committee on Aging, Fraud Hotline, 1-855-303-9470, https://www.aging.senate.gov/fraud-hotline. The FBI also directs victims age 60 or older who need help filing an IC3 complaint to the Department of Justice Elder Justice Hotline at 1- 833-FRAUD-11.
* * *
Policy and Legislative Recommendations
We urge Congress and federal agencies to advance the following legislative and policy actions to strengthen the nation's ability to fight fraud and address the rapid rise in AI-enabled scams:
1. Establish a National Office for Scam and Fraud Prevention
ABA proposes legislation to establish a National Office for Scam and Fraud Prevention within the Executive Office of the President, led by a Senate-confirmed Director. The Office would develop and coordinate a National Scam and Fraud Prevention and Response Strategy, align federal agencies, promote real-time public-private information sharing, improve national fraud data, coordinate public education, support state and local capacity, strengthen international cooperation, and review relevant federal budgets.
This legislation is needed because fraud crosses agency, industry, jurisdictional, and national boundaries, while federal responsibility remains fragmented. No single official currently has the authority and accountability to coordinate prevention across the entire ecosystem. The proposed Office would shift the federal response from a collection of agency programs to a coordinated national strategy focused on prevention, disruption, victim support, and measurable outcomes.
2. Enact a Comprehensive Anti-Fraud Telecom Package
ABA has developed a five-part legislative package to address persistent weaknesses in the telecommunications system. The package is intended to prevent criminals from obtaining access to trusted communications infrastructure, improve the reliability of caller identity information, and enable faster intervention when an active campaign is underway.
Part I: Performance Bond
The first title would require providers filing in the Robocall Mitigation Database to post a performance bond of up to $100,000 unless the Federal Communications Commission (FCC) determines that a bond is unnecessary. Bad-actor shell providers can enter the market, facilitate large illegal calling campaigns, incur penalties, and disappear without meaningful assets. A bond would create a barrier to entry for fraudulent providers and a source of recovery when penalties are imposed.
Part II: Rapid Response to Active Threat Campaigns The second title would establish an FCC rapid-response process for major, ongoing spoofing campaigns, with an objective of acting within 24 hours where practicable. It would authorize graduated actions, including formal traffic notifications, temporary cease-and-desist orders, and narrowly tailored blocking directives. It would also provide safe harbors for good-faith compliance and due-process protections for affected providers.
* * *
11 Morning Consult National Tracking Poll commissioned by the American Bankers Association, conducted Feb. 21-25, 2026, among 4,456 U.S. adults; margin of error +/- 1 percentage point.
* * *
This authority is needed because a high-volume campaign can cause substantial harm in hours or days, while traditional complaint and enforcement processes may take weeks or months. The government needs a mechanism capable of operating at the speed of scam.
Part III: Number Ownership Authentication and Attestation Integrity The third title would create a Number Ownership Authentication Database and a neutral governance framework. Before assigning the highest level of STIR/SHAKEN attestation, an originating provider would be required to verify that the caller has the lawful right to use the number displayed to the recipient.
STIR/SHAKEN can help ensure that caller identity information is transmitted without alteration, but the framework cannot protect consumers if false information is improperly authenticated at the beginning of the call chain. ABA-supported analysis found substantial volumes of spoofed calls receiving A- or B-level attestation, demonstrating that the integrity of the system depends on meaningful verification by the originating provider.
Part IV: SIM Farm and SIM Box Prohibition The fourth title would prohibit the manufacture, supply, possession, or operation of SIM farms and SIM boxes, subject to narrow lawful exceptions. These devices allow criminals to use a series of SIM cards to place large numbers of calls or messages that appear to originate from different numbers, making traceback and blocking substantially more difficult.
Part V: Database of Reported Scam Messages The fifth title would create a secure database containing scam and spam messages reported by consumers through "Report Junk" and similar tools. Authorized businesses, banks, platforms, telecom providers, and anti-fraud organizations could use the data to identify active impersonation campaigns and warn consumers.
Today, messaging providers receive valuable reports about scam campaigns, but the businesses whose brands are being impersonated often cannot access that intelligence. A privacy-protective, near-real-time database would help transform isolated consumer reports into actionable warnings and mitigation.
3. Enact the SCAM Act
Social media and digital advertising platforms should verify advertisers, detect impersonation, provide free and accessible reporting channels, investigate reports rapidly, remove confirmed fraudulent content, and prevent repeat offenders from returning. Paid advertising deserves particular attention because platforms actively sell, target, curate, and profit from the content.
ABA strongly supports S. 3774, the Safeguarding Consumers from Advertising Misconduct Act, or SCAM Act. This bipartisan legislation would translate these expectations into enforceable obligations by requiring online platforms to verify advertisers' identities, maintain systems to detect impersonation, investigate reported fraudulent or deceptive advertisements within 72 hours, and notify the reporting individual, business, or government agency of the outcome within 24 hours after completing the investigation. By requiring meaningful prevention and timely action against fraudulent paid advertising, the SCAM Act would help stop scams before criminals can establish trust and persuade victims to send money. Congress should enact it without delay.
4. Enact the National Payment Fraud Crime Control Act
ABA proposes the National Payment Fraud Crime Control Act, which would establish a Bureau of Justice Assistance grant program to help states create or strengthen Financial Crimes Intelligence Centers. These centers would coordinate the prevention, investigation, and prosecution of payment fraud, including both unauthorized transactions and payments victims are deceived into authorizing. Funding could support specialized investigators and analysts, training, technical assistance, centralized reporting, and information sharing among law enforcement, financial institutions, payment networks, and other partners.
The proposal builds on the Texas Financial Crimes Intelligence Center, a statewide fusion center staffed by experienced law enforcement officers and analysts that coordinates organized financial-crime investigations and works with federal, state, and local agencies and private-sector partners. Utah has also approved funding for a statewide Financial Crimes Intelligence Center within its Attorney General's Office. These models recognize that federal agencies cannot investigate every case and that state and local authorities need specialized capabilities to respond to the speed, volume, and complexity of modern fraud.
5. Require Strong Telecom Know-Your-Customer Controls and Enforcement
ABA commends the FCC and Chairman Carr for advancing proposals to require voice service providers to take specific actions to keep criminals off calling networks, including proposing stronger "know their customer" requirements for originating providers. In addition, we urge the FCC to require originating providers to collect and verify meaningful information about business callers before allowing them to use calling networks. Requirements should include legal identity, physical address, corporate formation and good-standing records, registration information, commercial presence, intended calling purpose, relevant ownership and compliance history, and evidence of financial responsibility.
ABA-supported analysis identified an originating provider that went from no calls to more than 136 million calls in a single month within two months of opening, with analysis indicating that most of the traffic was illegal. This example demonstrates why flexible and nonspecific expectations are inadequate. The FCC should establish clear standards, meaningful per-call penalties, allocate sufficient enforcement resources, and implement safeguards to prevent excluded bad actors from reentering the marketplace under new corporate identities.
6. Modernize Identity and Authentication
Policymakers should support phishing-resistant authentication, digital mobile driver's licenses and other verifiable digital credentials, privacy-preserving government identity-validation services, clear regulatory treatment of modern credentials, and improved remote identity-proofing standards. These measures provide a long-term structural response to both deepfake fraud and synthetic identity creation.
7. Improve Information Sharing, Intervention, and Recovery Congress should enact a new, fraud-specific information-sharing law rather than rely on further clarification of existing Bank Secrecy Act and anti-money-laundering authorities. FinCEN's June 2026 Section 314(b) fact sheet/12 is helpful in confirming that participating financial institutions may share fraud-related information in real time, but Section 314(b) remains a voluntary AML framework built around suspected money laundering or terrorist activity, with eligibility, registration, use, and confidentiality requirements. It was not designed to support the automated, cross-sector, and cross-border exchange of actionable intelligence needed at the speed of scam.
The current legal and technical environment also lacks the common data standards, operating rules, and shared infrastructure necessary to exchange and act on fraud signals in real time. A new law must address not only what information may be shared, but also who will build, own, operate, secure, fund, and oversee the network through which that information moves.
Congress should establish an explicit safe harbor from federal and state liability for entities that, in good faith, send, receive, and act on fraud-risk information, including by delaying or restraining suspicious funds, while maintaining appropriate privacy, security, governance, and redress protections. The framework should support near-real-time sharing among financial institutions, telecommunications providers, digital platforms, payment systems, law enforcement, and trusted intermediaries; strengthen bank-to-bank freeze and recovery processes; and enable interoperability among domestic and international fraud-information networks. This approach is consistent with the recent Royal United Services Institute Future of Financial Intelligence Sharing report's call for more connected, cross-sector and cross-border platforms operating as a "network of networks," rather than isolated information-sharing arrangements./13
8. Develop a Unified National Consumer Message Government agencies, banks, telecom providers, technology platforms, law enforcement, community organizations, and senior-serving groups should reinforce the same simple actions: pause, verify independently, use a family password, refuse secrecy, and call the bank quickly.
Conclusion
Generative AI is making traditional impersonation scams faster, cheaper, more personalized, and more convincing. Older Americans may suffer especially severe and irreversible consequences.
Banks are using advanced technology, employee intervention, education, information sharing, and recovery tools to protect customers, but banks cannot prevent fraud alone when the deception begins on a telecommunications network, social media platform, digital advertisement, messaging service, or fraudulent website.
Seniors should not be expected to distinguish, on their own and in real time, between a loved one and a synthetic voice, between a government official and an AI-generated video, or between their bank and a criminal using the bank's name and telephone number. The durable solution is to prevent criminals from exploiting the communications and technology systems that allow them to manufacture trust.
Congress can help by establishing accountable national leadership, strengthening telecommunications safeguards, improving information sharing and funds recovery, supporting modern identity systems, and ensuring that every sector involved in the scam lifecycle is responsible for protecting the public.
* * *
12 https://www.fincen.gov/system/files/shared/314bfactsheet.pdf
13 https://www.future-fis.com/fraudplatforms.html
* * *
Thank you for the opportunity to testify. I look forward to your questions.
* * *
Original text here: https://www.aging.senate.gov/imo/media/doc/32adb490-b498-1617-26d8-a24c018babde/Testimony_Benda%2007.29.26_fa521f44-1878-4a56-be7f-56ea5f26d53f.pdf
* * *
Chairman Scott, Ranking Member Gillibrand, and members of the Committee, thank you for the opportunity to testify for today's hearing entitled: "The AI Deception Machine: Deepfakes, Chatbots, and the New Frontier of Senior Fraud." My name is Paul Benda, and I serve ... Show Full Article WASHINGTON, Aug. 19 -- The Senate Special Committee on Aging released the following testimony by Paul Benda, executive vice president for risk, fraud and cybersecurity at the American Bankers Association, from a July 29, 2026, hearing entitled "The AI Deception Machine: Deepfakes, Chatbots, and the New Frontier of Senior Fraud": * * * Chairman Scott, Ranking Member Gillibrand, and members of the Committee, thank you for the opportunity to testify for today's hearing entitled: "The AI Deception Machine: Deepfakes, Chatbots, and the New Frontier of Senior Fraud." My name is Paul Benda, and I serveas Executive Vice President for Risk, Fraud and Cybersecurity at the American Bankers Association. ABA is the voice of the nation's banking industry, which is composed of small, regional and large banks that together employ more than two million people, safeguard trillions of dollars in deposits, and extend credit in communities across the country.
Banks play a unique and essential role in supporting the financial health of older Americans. As trusted institutions deeply rooted in their communities, banks are often the first, and most consistent, financial touchpoint for seniors and their families. Every day, bank employees help older customers manage retirement income, protect their savings, recognize and avoid scams, and adapt to an increasingly digital financial system. These relationships position banks not only as providers of financial services, but also as front line educators and protectors of financial security.
Banks have long used advanced analytics and artificial intelligence to identify suspicious transactions, protect customer accounts, and respond to cyber threats. Generative AI is strengthening those defensive tools. At the same time, criminals are using the same technology to make familiar scams more convincing, more personalized, cheaper to launch, and easier to scale.
The central point of my testimony is straightforward: generative AI is not replacing traditional scams. It is industrializing them. A criminal can now create a convincing voice, video, photograph, text message, advertisement, or online persona with little technical skill and at very low cost. AI allows fraudsters to imitate a friend, family member, bank employee, government official, physician, pastor, financial adviser, or virtually anyone else whose identity may inspire trust.
Older Americans face especially serious consequences from these scams. A loss may represent retirement savings that cannot be replaced, proceeds from a home sale, or money set aside for medical and long-term care needs. The financial damage can be compounded by shame, isolation, fear of losing independence, and reluctance to report the crime. We should be careful not to describe older Americans as inherently unsophisticated. These crimes succeed because professional criminals deliberately exploit trust, fear, urgency, family relationships, and authority. Generative AI makes those tactics substantially more credible.
My testimony focuses on four issues. First, I will describe how generative AI is changing the nature of scams. Second, I will explain why those changes are particularly significant for older Americans. Third, I will discuss what banks and ABA are doing to protect customers. Finally, I will offer practical steps for consumers and targeted policy recommendations for Congress and federal agencies.
Fraud Has Become an International Criminal Industrial Complex Today's scams are often not isolated crimes committed by a lone opportunist. They are carried out by sophisticated domestic and transnational criminal networks that use stolen personal information, fake identities, money mules, fraudulent advertisements, spoofed calls and texts, fake websites, encrypted communications, and rapid payment channels. The infrastructure that supports these crimes already exists, and generative AI is now being layered onto that infrastructure, making it even more effective.
AI can help criminals identify likely victims, collect personal information from public sources and data breaches, create customized scripts, translate communications into multiple languages, and sustain conversations with many victims at once. It can generate polished messages without the grammatical errors or awkward phrasing that once served as warning signs of a scam. It can also create synthetic identity documents and automate efforts to open accounts or take over existing accounts.
A recently published joint report by the ABA, Better Identity Coalition, and the Financial Services Sector Coordinating Council (FSSCC), Mitigating AI-Powered Attacks Against Identity and Authentication,/1 identifies three broad AI-enabled attack vectors: deepfake-driven social engineering and impersonation, synthetic identity creation, and the use of AI agents as attack surrogates capable of automating account takeover and fraud activity. These threats affect both consumers and financial institutions, but for many older Americans the most immediate danger is the use of AI to make social engineering and impersonation scams more convincing, personalized, and difficult to detect.
How Generative AI Is Changing Scams Against Older Americans
The fundamental danger is that seeing or hearing someone can no longer be treated as proof of identity. A voice message may sound like a loved one. A video may appear to show a government official. A caller ID display may show the correct name and phone number of a bank. A social media account may use a familiar photograph and a convincing history of posts.
Any of these signs of authenticity can be fabricated or manipulated.
Consumers have traditionally been advised to look for poor grammar, unnatural speech, obvious image defects, or inconsistent stories. Those remain useful clues, but they are becoming less reliable. A scam does not need to be technically perfect. It only needs to be credible enough, particularly when combined with urgency, secrecy, stolen personal information, and emotional manipulation.
A criminal can use information from social media, public records, obituaries, data brokers, breached accounts, and prior scam reports to tailor an approach. The scammer may know the names of family members, details from a recent trip, a medical condition, the victim's bank, or the identity of a trusted professional. That information can be woven into an AI-generated message or conversation, turning a generic scam into a seemingly personal emergency.
Generative AI enables a single criminal organization to maintain many apparent relationships at the same time. It can produce messages around the clock, adapt responses based on what a victim says, and translate content rapidly. This capability is especially dangerous in romance, investment, and recovery scams, where criminals may spend weeks or months building trust with their targets before requesting money.
A single fraud may begin with a social media advertisement, continue through text messages, move to a phone or video call, direct the victim to a fake website, and end with a bank transfer, cryptocurrency purchase, cash withdrawal, or courier pickup. Each channel can appear to confirm the others even when the entire experience is controlled by the same criminal group.
* * *
1 Financial Services Sector Coordinating Council, Mitigating AI-Powered Attacks Against Identity and Authentication (February 2026), https://fsscc.org/wp-content/uploads/2026/02/AI-IA-Workstream-Mitigations.pdf
* * *
A Recent FBI Warning Shows the Threat in Practice
A July 20, 2026, public service announcement from the FBI's Internet Crime Complaint Center provides a timely example of how generative AI is being used to enhance and scale fraud schemes. The FBI warned that criminals are impersonating FBI and IC3 personnel to revictimize people who have already lost money. These scams use fake social media profiles, spoofed government websites, and AI-generated videos that depict senior FBI personnel encouraging victims to submit information through a fraudulent IC3 website./2
The alert is particularly important because it incorporates many of the defining features of this emerging threat. Criminals target prior victims, impersonate trusted authorities, use AI-generated videos to create a false sense of legitimacy, distribute content through social media, and direct victims to websites designed to collect personal and financial information. The promise of recovering stolen funds is itself a scam designed to steal even more money.
The FBI also warns that scammers may use AI-generated video during real-time chats to impersonate executives, law enforcement officials, or other authority figures. This illustrates why consumers cannot rely solely on the apparent realism of a video or voice. The correct response is independent verification through a trusted channel.
Scam Scenarios Particularly Relevant to Older Americans
Grandchild and family-emergency scams - A victim receives a call that sounds like a child, grandchild, or other relative in distress. The caller claims to have been arrested, injured, kidnapped, or hospitalized. A second person may pose as a lawyer, police officer, doctor, or court official. The victim is instructed to send money immediately and not contact other family members. Voice cloning can remove one of the most important reasons a victim might otherwise question the call.
Government and law enforcement impersonation - A criminal may claim to represent the FBI, Social Security Administration, IRS, Medicare, a local sheriff, or another agency. The victim may be told that benefits will be suspended, that an arrest is imminent, that an identity has been compromised, or that money must be transferred for safekeeping. The same tactic may be used in recovery scams, where the criminal falsely promises to recover an earlier loss in exchange for a fee or additional payment.
Bank-impersonation scams - A victim receives a text or call that appears to come from the bank's fraud department. The caller may know partial account information, refer to a supposed transaction, or cause the bank's actual number to appear on caller ID. The criminal may ask the victim to disclose an authentication code, approve a login, install remote-access software, or move funds to a supposed "safe account." The FCC reinforced this warning in a July 21, 2026, consumer alert,/3 noting that bank-impersonation scams account for the highest losses of any impersonation scam category and that victims may lose everything in their accounts. The FCC emphasized a simple rule: banks will never call or text consumers and ask them to move money to protect it.
* * *
2 Federal Bureau of Investigation, Internet Crime Complaint Center, "FBI Warns of Scammers Impersonating the IC3," Alert No. I- 072026-PSA (July 20, 2026), https://www.ic3.gov/PSA/2026/PSA260720.
3 Federal Communications Commission, FCC Consumer Alert: Bank Impersonation Scams (July 21, 2026).
* * *
A recent bank survey illustrates the scale of the problem. Among 14 surveyed large banks, the average number of identified bank-impersonation scams increased 150 percent from 2024 to 2025, reaching an average of 26,196 scams per bank. Telecom-originated scams increased 124 percent, while scams originating on social media more than doubled.
The survey is not a census of the entire banking industry, but it provides strong directional evidence from institutions representing a significant share of industry assets./4
The same survey found that social media companies took nearly two weeks on average to remove bank-impersonation content and did not respond to more than 20 percent of takedown requests. Two surveyed banks reported that 40 percent of calls appearing to originate from bank-owned numbers had been spoofed. These results underscore that banks cannot stop impersonation scams alone. The deception often reaches the consumer through a telecom network, social media platform, advertising system, or messaging service before a bank sees any transaction./5
Romance and investment scams - AI-generated photographs, voice messages, video chats, and chatbots can help criminals build and sustain false relationships over long periods of time. A single criminal organization may manage many such relationships simultaneously. The relationship may eventually be used to persuade a victim to invest in a fraudulent scheme, send money in response to a manufactured emergency, receive and transfer funds, or participate in another form of financial exploitation. The emotional bond can make intervention extraordinarily difficult, even when family members or bank employees identify clear warning signs.
Technical-support and account-security scams - A criminal may impersonate a technology company, bank security specialist, or government cybersecurity official and claim that the victim's device or account has been compromised. The victim may be directed to install remote-access software, share a screen, reveal an authentication code, or transfer funds. AI-generated websites, instructions, and voices can make the approach look professional and legitimate.
Recovery and repeat-victimization scams - Prior victims are often targeted again by people claiming to be lawyers, investigators, banks, cybersecurity firms, or recovery services. The victim is promised that stolen funds have been located but must first pay a fee, tax, or processing charge. The recent FBI alert specifically warns about this form of "double-dip" scam. It is especially harmful because a prior victim may be emotionally vulnerable and desperate to recover retirement savings.
Why the Consequences for Older Americans Can Be Especially Severe The financial impact of a scam can be devastating at any age, but older victims may have less time and fewer employment opportunities to rebuild savings. Losses may involve retirement assets, home equity, emergency funds, or money needed for health and long-term care.
* * *
4 Tara Payne, Drew Ruben & Madison Stulga, Bank Policy Institute, "Hijacking Trust: How Scammers Exploit the Banking System's Most Valuable Asset" (July 20, 2026), https://bpi.com/impersonation/. The survey covered 14 BPI member banks ranging from approximately $100 billion to more than $1 trillion in assets.
5 Id.
* * *
Victims may also hesitate to report. Some fear embarrassment or worry that family members will question their ability to manage their own finances. Others remain under the criminal's influence and believe that one more payment will resolve the emergency or recover prior losses. In romance scams, the victim may remain emotionally attached to the person they believe is real.
These dynamics make scam intervention different from ordinary transaction monitoring. Bank employees sometimes encounter customers who have been coached to lie about the purpose of a transaction, conceal communications, or distrust the very bank employee trying to protect them.
Effective intervention therefore requires empathy, training, and an understanding of the psychological manipulation used by professional criminals.
How Banks Are Protecting Customers
Banks employ a host of tools to detect and deter fraud, including AI, machine learning, and other advanced analytics to identify unusual transactions, new or suspicious payees, account takeover, abnormal device activity, behavioral anomalies, manipulated documents, and high-risk payment patterns. These systems can surface transactions for additional review or prompt the bank to contact the customer before a transaction is completed.
While criminals use AI to make deception more effective, banks employ AI to detect anomalies and protect customers. The goal should not be to restrict beneficial defensive uses. It should be to strengthen safeguards, improve information sharing, and reduce criminals' ability to exploit communications and identity systems.
Banks increasingly employ layered defenses that combine identity validation, device and behavioral signals, document analysis, liveness and injection-attack detection, out-of-band verification, additional confirmation for high-risk activity, and human review. There is no single "deepfake detector" that can solve the problem. Effective defense depends on multiple independent signals and risk-based controls.
Technology alone, however, cannot identify every scam. Banks also rely on human judgment because no model, algorithm, or automated system can reliably detect every form of deception.
Front-line bank employees are trained to recognize sudden payments to new recipients, unusual withdrawals, customers acting under extreme urgency, demands for secrecy, customers being coached over the telephone, and requests involving "safe accounts," cryptocurrency, gift cards, cash couriers, or other unusual payment methods.
ABA is expanding training that helps bankers recognize deepfakes and AI-enabled threats, support victims of financial crime, and intervene when a customer appears to be under a scammer's influence. That includes ABA's Spot the Scammer training and work focused on "breaking the scam spell," which recognizes that the challenge is not merely identifying an unusual transaction but helping a victim step outside the criminal's narrative./6
Fraud alerts and direct bank-customer communications are among the most effective tools available. A February 2026 Morning Consult poll commissioned by ABA found that nearly nine in ten bank customers said their bank takes proactive steps to protect them from fraud and scams.
* * *
6 American Bankers Association, "Spot the Scammer: How to Detect Scams, Deepfakes, and AI Threats," https://www.aba.com/training-events/online-training/spot-the-scammer.
* * *
Three in five consumers had received a fraud alert regarding potentially suspicious account activity, and 96 percent of those recipients found the alerts valuable./7
The same poll found that 72 percent of consumers believe banks do more than other industries to protect them from fraud and scams. By a margin of nearly five to one, consumers trusted banks more than other industries and the government to protect them from fraud. These findings are important because proposed legislation and the regulatory environment should support, rather than inadvertently impede, timely fraud alerts and other protective communications./8
Consumer education is an equally important component of fraud prevention. The ABA Foundation's Safe Banking for Seniors program provides banks with free, turnkey materials to educate older Americans, caregivers, and families about scams, identity theft, financial caregiving, and financial exploitation. More than 2,550 banks have participated in the program since it was launched in 2016.
ABA and the FBI have also developed consumer materials specifically addressing deepfake scams, and ABA maintains guidance explaining how deepfake media can be used in impersonation, investment, romance, employment, and other fraud schemes. ABA's #BanksNeverAskThat campaign teaches consumers to recognize common impersonation tactics, while the current "Snap Out of It!" theme focuses on interrupting the urgency and psychological manipulation used by scammers./9
ABA and the banking industry are working to improve the speed of bank-to-bank communications, standardize documentation used to request holds or returns of fraudulent funds, expand the ABA Fraud Contact Directory, and develop mechanisms for sharing fraud indicators.
ABA is also working with law enforcement and other partners to improve recovery of funds, including funds moved through cryptocurrency.
These efforts are critical because modern fraud schemes span multiple institutions and industries.
One bank may see the victim's outgoing payment, another may hold the receiving account, a telecom provider may carry the spoofed call, and a platform may host the fraudulent advertisement. No participant has the complete picture. Faster, privacy-protective sharing of fraud signals can allow earlier intervention and improve the chance of recovery.
What Older Americans and Their Families Can Do Consumer guidance should be memorable, focused on the mechanics of impersonation, and avoid requiring people to become deepfake experts. The goal is to change the decision process so that a convincing voice or video cannot by itself trigger a payment. ABA recommends that seniors and their families take the following steps to guard against fraud: Pause - Do not act immediately, no matter how urgent or frightening the communication appears. Time is one of the most effective defenses against an AI-enabled scam because pausing interrupts the criminal's script and creates an opportunity to verify the story.
* * *
7 Morning Consult National Tracking Poll commissioned by the American Bankers Association, conducted Feb. 21-25, 2026, among 4,456 U.S. adults; margin of error +/- 1 percentage point.
8 Id.
9 American Bankers Association, "Deepfake Media Scams," https://www.aba.com/news-research/analysis-guides/deepfake-media- scams; American Bankers Association, "ABA Foundation and FBI Release New Infographic to Help Americans Spot and Avoid Deepfake Scams," https://www.aba.com/about-us/press-room/press-releases/ABA-Foundation-and-FBI-Joint-Infographic-on- Deepfake-Scams.
* * *
Verify through a separate channel - Hang up and contact the person or organization using a telephone number or method already known to be legitimate. Do not use a callback number, link, or website supplied by the caller or message. Call another family member before sending money. Contact the bank using the number printed on the card or statement.
Use a family password - Families should agree in advance on a private word or phrase that can be used to verify a genuine emergency. The phrase should not be posted online or easily guessed. A family can also establish a personal question whose answer is not available through social media or public records.
Remember: "If it's a secret, it's a scam" - A demand to act now, stay on the line, or keep the matter secret is a powerful warning sign. Criminals use secrecy to prevent victims from contacting family, banks, or law enforcement. A legitimate bank, government agency, police department, or family member should not object to independent verification.
Do not trust the channel itself - Caller ID can be spoofed. Voices can be cloned.
Photographs and videos can be fabricated. Social media accounts can be copied or compromised. Seeing a familiar face or hearing a familiar voice is no longer sufficient proof of identity.
Never move money to a "safe account" - A bank or government agency will not direct a consumer to transfer funds to another account for protection. Requests involving cryptocurrency, gift cards, cash couriers, or payments to newly created accounts deserve particular scrutiny.
Involve a trusted person and report quickly - Before making a large or unusual payment, speak with a trusted family member, friend, banker, caregiver, attorney, or financial adviser. If money has already been sent, contact the bank immediately. Rapid reporting may improve the chance of stopping or recovering funds. Victims should also report the incident to the FBI's Internet Crime Complaint Center. The Senate Special Committee on Aging operates a Fraud Hotline that can help older Americans and families understand how to report scams and protect themselves./10
Most importantly, victims should not be shamed. These schemes are designed by professional criminals using sophisticated technology and psychological manipulation. Reporting quickly helps the victim and may help prevent others from being harmed.
Banks and Consumers Cannot Solve This Alone By the time a bank sees a transaction, a criminal may have communicated with the victim for days or months. The deception may have been enabled by a spoofed call, fraudulent text, paid advertisement, social media account, messaging application, fake website, or cryptocurrency platform. Responsibility must extend to the sectors that enable criminals to reach and manipulate victims.
The Morning Consult poll referenced above found that three in four Americans had seen fraudulent or deceptive advertisements on social media. Seventy-eight percent supported federal legislation requiring social media companies to do more to identify and remove fake accounts and fraudulent advertisements, and 77 percent supported regulatory action requiring telecommunications providers to do more to prevent spoofed caller IDs. Consumers clearly do not view scam prevention as solely a bank responsibility./11
* * *
10 U.S. Senate Special Committee on Aging, Fraud Hotline, 1-855-303-9470, https://www.aging.senate.gov/fraud-hotline. The FBI also directs victims age 60 or older who need help filing an IC3 complaint to the Department of Justice Elder Justice Hotline at 1- 833-FRAUD-11.
* * *
Policy and Legislative Recommendations
We urge Congress and federal agencies to advance the following legislative and policy actions to strengthen the nation's ability to fight fraud and address the rapid rise in AI-enabled scams:
1. Establish a National Office for Scam and Fraud Prevention
ABA proposes legislation to establish a National Office for Scam and Fraud Prevention within the Executive Office of the President, led by a Senate-confirmed Director. The Office would develop and coordinate a National Scam and Fraud Prevention and Response Strategy, align federal agencies, promote real-time public-private information sharing, improve national fraud data, coordinate public education, support state and local capacity, strengthen international cooperation, and review relevant federal budgets.
This legislation is needed because fraud crosses agency, industry, jurisdictional, and national boundaries, while federal responsibility remains fragmented. No single official currently has the authority and accountability to coordinate prevention across the entire ecosystem. The proposed Office would shift the federal response from a collection of agency programs to a coordinated national strategy focused on prevention, disruption, victim support, and measurable outcomes.
2. Enact a Comprehensive Anti-Fraud Telecom Package
ABA has developed a five-part legislative package to address persistent weaknesses in the telecommunications system. The package is intended to prevent criminals from obtaining access to trusted communications infrastructure, improve the reliability of caller identity information, and enable faster intervention when an active campaign is underway.
Part I: Performance Bond
The first title would require providers filing in the Robocall Mitigation Database to post a performance bond of up to $100,000 unless the Federal Communications Commission (FCC) determines that a bond is unnecessary. Bad-actor shell providers can enter the market, facilitate large illegal calling campaigns, incur penalties, and disappear without meaningful assets. A bond would create a barrier to entry for fraudulent providers and a source of recovery when penalties are imposed.
Part II: Rapid Response to Active Threat Campaigns The second title would establish an FCC rapid-response process for major, ongoing spoofing campaigns, with an objective of acting within 24 hours where practicable. It would authorize graduated actions, including formal traffic notifications, temporary cease-and-desist orders, and narrowly tailored blocking directives. It would also provide safe harbors for good-faith compliance and due-process protections for affected providers.
* * *
11 Morning Consult National Tracking Poll commissioned by the American Bankers Association, conducted Feb. 21-25, 2026, among 4,456 U.S. adults; margin of error +/- 1 percentage point.
* * *
This authority is needed because a high-volume campaign can cause substantial harm in hours or days, while traditional complaint and enforcement processes may take weeks or months. The government needs a mechanism capable of operating at the speed of scam.
Part III: Number Ownership Authentication and Attestation Integrity The third title would create a Number Ownership Authentication Database and a neutral governance framework. Before assigning the highest level of STIR/SHAKEN attestation, an originating provider would be required to verify that the caller has the lawful right to use the number displayed to the recipient.
STIR/SHAKEN can help ensure that caller identity information is transmitted without alteration, but the framework cannot protect consumers if false information is improperly authenticated at the beginning of the call chain. ABA-supported analysis found substantial volumes of spoofed calls receiving A- or B-level attestation, demonstrating that the integrity of the system depends on meaningful verification by the originating provider.
Part IV: SIM Farm and SIM Box Prohibition The fourth title would prohibit the manufacture, supply, possession, or operation of SIM farms and SIM boxes, subject to narrow lawful exceptions. These devices allow criminals to use a series of SIM cards to place large numbers of calls or messages that appear to originate from different numbers, making traceback and blocking substantially more difficult.
Part V: Database of Reported Scam Messages The fifth title would create a secure database containing scam and spam messages reported by consumers through "Report Junk" and similar tools. Authorized businesses, banks, platforms, telecom providers, and anti-fraud organizations could use the data to identify active impersonation campaigns and warn consumers.
Today, messaging providers receive valuable reports about scam campaigns, but the businesses whose brands are being impersonated often cannot access that intelligence. A privacy-protective, near-real-time database would help transform isolated consumer reports into actionable warnings and mitigation.
3. Enact the SCAM Act
Social media and digital advertising platforms should verify advertisers, detect impersonation, provide free and accessible reporting channels, investigate reports rapidly, remove confirmed fraudulent content, and prevent repeat offenders from returning. Paid advertising deserves particular attention because platforms actively sell, target, curate, and profit from the content.
ABA strongly supports S. 3774, the Safeguarding Consumers from Advertising Misconduct Act, or SCAM Act. This bipartisan legislation would translate these expectations into enforceable obligations by requiring online platforms to verify advertisers' identities, maintain systems to detect impersonation, investigate reported fraudulent or deceptive advertisements within 72 hours, and notify the reporting individual, business, or government agency of the outcome within 24 hours after completing the investigation. By requiring meaningful prevention and timely action against fraudulent paid advertising, the SCAM Act would help stop scams before criminals can establish trust and persuade victims to send money. Congress should enact it without delay.
4. Enact the National Payment Fraud Crime Control Act
ABA proposes the National Payment Fraud Crime Control Act, which would establish a Bureau of Justice Assistance grant program to help states create or strengthen Financial Crimes Intelligence Centers. These centers would coordinate the prevention, investigation, and prosecution of payment fraud, including both unauthorized transactions and payments victims are deceived into authorizing. Funding could support specialized investigators and analysts, training, technical assistance, centralized reporting, and information sharing among law enforcement, financial institutions, payment networks, and other partners.
The proposal builds on the Texas Financial Crimes Intelligence Center, a statewide fusion center staffed by experienced law enforcement officers and analysts that coordinates organized financial-crime investigations and works with federal, state, and local agencies and private-sector partners. Utah has also approved funding for a statewide Financial Crimes Intelligence Center within its Attorney General's Office. These models recognize that federal agencies cannot investigate every case and that state and local authorities need specialized capabilities to respond to the speed, volume, and complexity of modern fraud.
5. Require Strong Telecom Know-Your-Customer Controls and Enforcement
ABA commends the FCC and Chairman Carr for advancing proposals to require voice service providers to take specific actions to keep criminals off calling networks, including proposing stronger "know their customer" requirements for originating providers. In addition, we urge the FCC to require originating providers to collect and verify meaningful information about business callers before allowing them to use calling networks. Requirements should include legal identity, physical address, corporate formation and good-standing records, registration information, commercial presence, intended calling purpose, relevant ownership and compliance history, and evidence of financial responsibility.
ABA-supported analysis identified an originating provider that went from no calls to more than 136 million calls in a single month within two months of opening, with analysis indicating that most of the traffic was illegal. This example demonstrates why flexible and nonspecific expectations are inadequate. The FCC should establish clear standards, meaningful per-call penalties, allocate sufficient enforcement resources, and implement safeguards to prevent excluded bad actors from reentering the marketplace under new corporate identities.
6. Modernize Identity and Authentication
Policymakers should support phishing-resistant authentication, digital mobile driver's licenses and other verifiable digital credentials, privacy-preserving government identity-validation services, clear regulatory treatment of modern credentials, and improved remote identity-proofing standards. These measures provide a long-term structural response to both deepfake fraud and synthetic identity creation.
7. Improve Information Sharing, Intervention, and Recovery Congress should enact a new, fraud-specific information-sharing law rather than rely on further clarification of existing Bank Secrecy Act and anti-money-laundering authorities. FinCEN's June 2026 Section 314(b) fact sheet/12 is helpful in confirming that participating financial institutions may share fraud-related information in real time, but Section 314(b) remains a voluntary AML framework built around suspected money laundering or terrorist activity, with eligibility, registration, use, and confidentiality requirements. It was not designed to support the automated, cross-sector, and cross-border exchange of actionable intelligence needed at the speed of scam.
The current legal and technical environment also lacks the common data standards, operating rules, and shared infrastructure necessary to exchange and act on fraud signals in real time. A new law must address not only what information may be shared, but also who will build, own, operate, secure, fund, and oversee the network through which that information moves.
Congress should establish an explicit safe harbor from federal and state liability for entities that, in good faith, send, receive, and act on fraud-risk information, including by delaying or restraining suspicious funds, while maintaining appropriate privacy, security, governance, and redress protections. The framework should support near-real-time sharing among financial institutions, telecommunications providers, digital platforms, payment systems, law enforcement, and trusted intermediaries; strengthen bank-to-bank freeze and recovery processes; and enable interoperability among domestic and international fraud-information networks. This approach is consistent with the recent Royal United Services Institute Future of Financial Intelligence Sharing report's call for more connected, cross-sector and cross-border platforms operating as a "network of networks," rather than isolated information-sharing arrangements./13
8. Develop a Unified National Consumer Message Government agencies, banks, telecom providers, technology platforms, law enforcement, community organizations, and senior-serving groups should reinforce the same simple actions: pause, verify independently, use a family password, refuse secrecy, and call the bank quickly.
Conclusion
Generative AI is making traditional impersonation scams faster, cheaper, more personalized, and more convincing. Older Americans may suffer especially severe and irreversible consequences.
Banks are using advanced technology, employee intervention, education, information sharing, and recovery tools to protect customers, but banks cannot prevent fraud alone when the deception begins on a telecommunications network, social media platform, digital advertisement, messaging service, or fraudulent website.
Seniors should not be expected to distinguish, on their own and in real time, between a loved one and a synthetic voice, between a government official and an AI-generated video, or between their bank and a criminal using the bank's name and telephone number. The durable solution is to prevent criminals from exploiting the communications and technology systems that allow them to manufacture trust.
Congress can help by establishing accountable national leadership, strengthening telecommunications safeguards, improving information sharing and funds recovery, supporting modern identity systems, and ensuring that every sector involved in the scam lifecycle is responsible for protecting the public.
* * *
12 https://www.fincen.gov/system/files/shared/314bfactsheet.pdf
13 https://www.future-fis.com/fraudplatforms.html
* * *
Thank you for the opportunity to testify. I look forward to your questions.
* * *
Original text here: https://www.aging.senate.gov/imo/media/doc/32adb490-b498-1617-26d8-a24c018babde/Testimony_Benda%2007.29.26_fa521f44-1878-4a56-be7f-56ea5f26d53f.pdf
Better Business Bureau of Southern Colorado CEO Liebert Testifies Before Senate Health, Education, Labor & Pensions Subcommittee
WASHINGTON, Aug. 18 -- The Senate Health, Education, Labor and Pensions Subcommittee on Employment and Workplace Safety released the following written testimony by Jonathan Liebert, CEO of Better Business Bureau of Southern Colorado, Colorado Springs, from a July 29, 2026, hearing entitled "The Impact of AI on the Workforce":
* * *
Artificial intelligence is already changing how Americans work, learn, make decisions, and compete. The central workforce challenge is no longer whether AI will be adopted; it is whether workers, employers, educators, and communities will be prepared to use it responsibly ... Show Full Article WASHINGTON, Aug. 18 -- The Senate Health, Education, Labor and Pensions Subcommittee on Employment and Workplace Safety released the following written testimony by Jonathan Liebert, CEO of Better Business Bureau of Southern Colorado, Colorado Springs, from a July 29, 2026, hearing entitled "The Impact of AI on the Workforce": * * * Artificial intelligence is already changing how Americans work, learn, make decisions, and compete. The central workforce challenge is no longer whether AI will be adopted; it is whether workers, employers, educators, and communities will be prepared to use it responsiblyand share in its benefits.
"AI is not an Oracle; it is an Ideation Engine"
Since 2025, I have trained over 5,500 people across Colorado and around the country on the practical and responsible use of AI. That work has given me a front-row seat to extraordinary curiosity and real anxiety.
Business owners want to know which tools are credible. Nonprofit leaders want to know how to expand impact without compromising privacy. Educators want to know how to prepare students for jobs that are changing faster than curricula. Workers want to know whether AI will replace them or help them become more capable.
Better Business Bureau of Southern Colorado created the BBB AI Hub (visit it here at www.bbbaihub.org) to help close the gap between the speed of technological change and the ability of people and organizations to respond. In collaboration with the Pikes Peak Workforce Center, the UCCS College of Business, Exponential Impact, the Colorado Springs Chamber & EDC, the Pikes Peak Small Business Development Center, the City of Colorado Springs, the Colorado AI Alliance, the Pikes Peak Business Alliance (PPBEA), and the National Association of Workforce Boards, the Hub connects AI literacy, responsible adoption, governance, and workforce preparation. Our experience suggests that the nation does not need to build an entirely new delivery system. Trusted local infrastructure already exists in workforce centers, community colleges, libraries, BBBs, chambers, and nonprofit networks. It needs practical curriculum, capacity, coordination, and sustained support.
My recommendations are straightforward: invest in rapid AI literacy through trusted local intermediaries; provide practical governance and vendor-evaluation tools for small organizations; improve task-level workforce data; prepare educators and students for AI-enabled work; protect human accountability in consequential employment decisions; and ensure rural communities, small employers, nonprofits, and workers outside technology centers can participate in the opportunity rather than absorb only the disruption.
I. A Cautiously Optimistic View of AI
My position is neither uncritically pro-AI nor reflexively anti-AI. The loudest public voices often occupy the extremes. One side presents AI as an unqualified good that should be adopted as quickly as possible. The other presents it as a force that will inevitably destroy jobs, erode learning, and diminish human value.
Neither perspective is useful to the small-business owner deciding whether to purchase a new platform, the workforce professional helping a displaced worker, the teacher redesigning an assignment, or the nonprofit director with three employees and no technology department.
Cautious optimism means holding two realities at the same time. AI can improve productivity, reduce administrative burden, strengthen customer service, broaden access to expertise, and help small organizations accomplish work that previously required larger teams. I have watched people use AI to organize complex information, improve communication, accelerate research, and turn ideas into usable plans.
I have also watched people enter sensitive customer information into public systems, accept fabricated citations as fact, rely on incomplete or biased outputs, and purchase products marketed as "AI-powered" without understanding what the technology actually did. These are not abstract concerns. They are predictable consequences of adoption moving faster than education and governance.
The question before us is not simply whether AI is good or bad. The better questions are: Who will understand it? Who will have access to useful training? Who will be accountable when it influences a consequential decision? Which communities will receive the productivity gains, and which will bear the disruption? Public policy should help answer those questions before market forces answer them by default.
II. What I Am Seeing in Classrooms and Communities Since 2025, I have trained over 5,500 individuals through classes, workshops, presentations, and community conversations across Colorado and around the country. The audiences have included business owners, employees, educators, nonprofit leaders, workforce professionals, students, and community leaders.
Although their industries and responsibilities differ, the same themes appear repeatedly.
* People are interested in AI, but many do not know where to begin or which information to trust.
* Many organizations are already using AI informally, even when leadership believes adoption has not started.
* Employees frequently receive access to tools before receiving guidance about privacy, verification, intellectual property, bias, or human review.
* Small organizations are overwhelmed by vendor claims and lack the legal, technical, and procurement capacity available to large enterprises.
* Educators recognize that AI will be part of students' futures, but they need support to distinguish responsible learning from shortcut-taking and academic dishonesty.
* Fear often declines when people receive hands-on instruction and a realistic understanding of both capabilities and limitations.
That is the posture I encourage in every room: curiosity about AI produces better questions, better verification, and better decisions than either uncritical enthusiasm or reflexive dismissal.
The most common problem I observe is not that people refuse to use AI. It is that they trust it too quickly or use it without enough context. AI can become a cognitive crutch that replaces thought, or a catalyst that improves thought. The difference is not the tool alone; it is the quality of instruction, the expectations around its use, and the presence of human judgment.
The framing I return to most often in class is this: AI is not an oracle, it is an ideation engine. An oracle is consulted for answers and believed. An ideation engine is used to generate options, surface considerations a person had not thought of, stress-test an assumption, draft a first version, or restate a complex idea in plainer terms. The output is raw material for thinking, not a conclusion to be adopted. When people treat AI as an oracle, they outsource judgment and inherit whatever errors, gaps, or biases the system carried. When they treat it as an ideation engine, they stay in the decision, and the technology makes them faster and more thorough without making them less accountable.
This distinction has practical consequences in the workplace. A manager who asks AI to draft three approaches to a staffing problem, then applies experience and knowledge of the team to choose among them, has used the tool well. A manager who asks AI who to lay off and acts on the answer has not. The same is true of a nurse checking a care protocol, a small-business owner reviewing a contract summary, or a teacher building a lesson plan. In every case the value comes from the human bringing context the system does not have--local knowledge, ethical judgment, and accountability for the outcome.
III. The BBB AI Hub: A Local Model for Workforce Preparation Better Business
Bureau has spent more than a century helping people make informed marketplace decisions and helping businesses build trust. AI is now changing both sides of that mission. It is changing how companies operate, how consumers encounter information, how products are marketed, and how fraud and misinformation can be created and scaled. For BBB of Southern Colorado, AI literacy and AI governance are therefore a natural extension of marketplace trust.
We created the BBB AI Hub to provide practical, accessible, and responsible AI education. The Hub is designed for people who do not need to become engineers but do need to make better decisions about AI. Its work centers on four connected priorities:
* * *
Priority ... What It Means
AI literacy ... Understanding what AI can and cannot do; writing effective instructions; verifying outputs; protecting sensitive information; and recognizing misinformation, bias, and fabricated content.
Workforce preparation ... Helping workers use AI to become more capable, helping employers redesign work thoughtfully, and connecting training to actual roles and advancement opportunities.
AI governance ... Creating clear rules about approved tools, restricted data, human review, accountability, documentation, and employee training.
Marketplace trust ... Helping organizations evaluate AI vendors and claims before making purchasing or implementation decisions.
* * *
Our collaboration with the Pikes Peak Workforce Center is especially important because AI education and workforce services are too often treated as separate activities. They should be connected. Workforce centers understand local employers, job seekers, occupations, and barriers to employment. Community-based organizations understand trust and access. Education providers understand learning. When these capabilities are aligned, AI training can be connected to real jobs, incumbent-worker development, reemployment, and advancement rather than existing as a stand-alone technology demonstration.
This partnership also offers a broader lesson for federal policy: the country already has local institutions capable of reaching workers and small employers. Federal support should strengthen these institutions rather than creating programs that are difficult to access, overly technical, or disconnected from local labor-market needs.
IV. AI Literacy Must Become a Foundational Workforce Skill AI literacy should not be confused with learning one product. Tools will change. Interfaces will change. The durable skill is knowing how to work with AI systems critically and responsibly. At a minimum, workers and employers should understand:
** How to describe a task clearly and provide appropriate context.
* How to evaluate whether AI is suitable for the task at all.
* How to verify factual claims, calculations, citations, and recommendations.
* How to protect customer, employee, student, patient, financial, and proprietary information.
* How bias can enter data, prompts, outputs, and decisions.
* When human review is required and who remains accountable.
* How to document and communicate the use of AI when transparency matters.
AI literacy should be role-based. A teacher, human-resources manager, small-business owner, customerservice employee, and cybersecurity professional do not need the same training. Organizations should identify the tasks people actually perform, the risks attached to those tasks, and the level of review required.
This is more effective than generic training and more realistic for employers with limited time and resources.
AI literacy is also inseparable from critical thinking. The presence of AI does not make critical thinking less important; it makes it more visible. A person must decide whether the question is well framed, whether the output is credible, what information is missing, and whether the result should influence a human decision.
Responsible AI use requires more judgment, not less.
V. AI Governance for Organizations of Every Size Many organizations are waiting for perfect certainty before establishing an AI policy. That is a mistake.
Governance does not require a large compliance department. It begins with practical decisions that any organization can make.
1. Adopt an AI use policy -- Define approved and prohibited uses, identify restricted information, and establish expectations for disclosure and human review.
2. Create a role-based literacy plan -- Train employees according to the decisions they make, the data they handle, and the risks associated with their work.
3. Require accountable human oversight -- AI may support a decision, but a named person should remain responsible when the decision affects employment, safety, education, finances, or access to services.
4. Assess and improve continuously -- Tools, risks, and regulations will change. Organizations should review their practices, incidents, and training on a regular schedule.
The goal of governance is not to prevent experimentation. It is to create a safe environment for experimentation. Clear boundaries give employees confidence about what they may do, protect the organization from avoidable risk, and help leadership learn from use rather than discover it only after a problem occurs.
VI. Vetting AI Businesses and Tools
The rapid growth of AI has created a marketplace challenge. "AI-powered" has no consistent practical meaning for a buyer. Some products use sophisticated models in meaningful ways. Others add a thin AI feature to an existing platform. Still others rely on marketing claims that a small employer cannot independently verify.
Large companies may have procurement teams, legal counsel, cybersecurity experts, data-governance professionals, and the leverage to negotiate contracts. Small employers and nonprofits frequently have none of these. They need plain-language standards and questions that help them evaluate risk before purchasing or connecting a tool to sensitive systems.
Questions every organization should ask an AI vendor
Area ... Questions to Ask
Data practices... What information is collected? Where is it stored? Is customer data used to train the system? Can data be deleted or exported?
Security ... What security controls and independent assessments are in place? How are incidents reported?
Performance ... What does the product actually do, and what evidence supports claims about accuracy, productivity, or cost savings?
Human accountability ... What happens when the system is wrong? Can a human review, override, and explain the result?
Bias and accessibility ... How has the system been evaluated across different users and populations?
Contract terms ... Who owns inputs and outputs? What happens to data when the contract ends? What indemnification or limitations apply?
Vendor credibility ... Who operates the company? Is support available? Are references and claims independently verifiable?
* * *
Trusted marketplace organizations can help translate technical questions into usable guidance. This is not a substitute for regulation, cybersecurity, or legal review. It is a practical layer of consumer and business education that can reduce harm before it occurs.
VII. Preparing Educators and Students for an AI-Enabled Workforce The workforce conversation cannot begin after graduation. Educators are being asked to prepare students for a labor market in which AI will be embedded across occupations, including roles that are not traditionally considered technical. At the same time, schools are navigating legitimate concerns about academic integrity, student privacy, unequal access, and overreliance on automated tools.
The answer is not to pretend AI is absent from the world students will enter. Nor is it to allow unrestricted use without expectations. Schools need age-appropriate AI literacy, clear policies, teacher professional development, and learning experiences that preserve foundational knowledge while teaching students how to use AI as a thought partner rather than a substitute for thought.
The most durable workforce skills will include communication, critical thinking, ethical judgment, creativity, adaptability, domain expertise, and the ability to evaluate AI-generated work. Students should learn how to ask better questions, challenge an output, identify missing context, and explain their own reasoning. These capabilities will help them whether they become nurses, entrepreneurs, technicians, teachers, public servants, or software developers.
VIII. Workforce Disruption, Opportunity, and Fairness Some jobs will be redesigned, and some tasks--particularly routine, entry-level, and administrative tasks-- will be reduced or automated. New roles will emerge, and existing workers may become more productive and capable. But disruption and opportunity are often the same event viewed from different sides.
The honest concern is that new opportunities may not appear in the same communities, at the same wages, or quickly enough for workers whose tasks are displaced. Productivity gains may accrue to organizations without being reflected in job quality, wages, advancement, or reduced workload. Rural communities and smaller employers may face the disruption without having equal access to training, infrastructure, or expertise.
Workers should have a meaningful voice in how AI is introduced into their jobs. Employers should communicate what is changing, what data is being used, how performance will be evaluated, and how employees can challenge an incorrect automated conclusion. Transparency and human accountability are especially important when AI materially affects hiring, promotion, scheduling, performance evaluation, discipline, layoffs, or workplace surveillance.
The message I bring to every class is practical: learn to use these tools strategically, ethically, and responsibly, and they can become leverage. They can help a worker perform the current job more effectively, preserve time for higher-value work, or compete for opportunities that were previously out of reach. That outcome is possible, but it is not automatic. Preparation determines who benefits.
IX. Recommendations to Congress
1. Fund rapid, practical AI upskilling -- Use trusted local intermediaries such as workforce centers, community colleges, libraries, BBBs, chambers, and nonprofit networks to deliver short-cycle, role-based training tied to employment and advancement outcomes.
2. Strengthen AI literacy for educators and students -- Support teacher professional development, age-appropriate curricula, privacy protections, and career-connected learning that prepares students to work responsibly with AI.
3. Equip small organizations with governance tools -- Develop and distribute model AI-use policies, vendor-evaluation guides, cybersecurity practices, and human-oversight frameworks that small businesses and nonprofits can realistically adopt.
4. Improve workforce intelligence -- Create timely, task-level, and geographically detailed data so policymakers and local workforce systems can identify disruption before layoffs and align training with changing employer demand.
5. Preserve transparency and human accountability -- When AI materially affects hiring, evaluation, discipline, layoffs, scheduling, or surveillance, workers should know it is being used and a person should remain accountable for the decision.
6. Ensure broad access to the benefits -- Prioritize rural communities, small employers, underserved populations, and workers outside major technology centers so they receive training and opportunity, not only disruption.
7. Evaluate outcomes, not just adoption -- Measure job quality, wage mobility, worker advancement, productivity, small-business competitiveness, safety, and access--not merely how many organizations purchased an AI tool.
Conclusion
AI itself is not the greatest workforce threat. The greater risk is allowing the technology to advance while workers, businesses, nonprofits, educators, and communities are left without the knowledge, protections, or support required to use it well.
America does not need to choose between innovation and responsibility. We can pursue both. We can help businesses compete while protecting workers. We can prepare students for AI-enabled careers while preserving the importance of human learning. We can encourage experimentation while requiring accountability. We can be optimistic about what AI makes possible and serious about the conditions required for those possibilities to benefit people broadly.
The future of work will not be predetermined by an algorithm. It will be shaped by the choices we make about who is educated, who is protected, who has access, who remains accountable, and who shares in the value that AI helps create.
Chairman Banks, Ranking Member Hickenlooper, and members of the Subcommittee, thank you for the opportunity to submit this testimony and for your attention to the workers, employers, educators, and communities navigating this transition.
* * *
Original text here: https://www.help.senate.gov/imo/media/doc/6cbbd241-b14f-d7fb-a82f-91145ad11dc1/Liebert%20Testimony_c03b7053-d741-46c3-a80a-442da4ed5114.pdf
* * *
Artificial intelligence is already changing how Americans work, learn, make decisions, and compete. The central workforce challenge is no longer whether AI will be adopted; it is whether workers, employers, educators, and communities will be prepared to use it responsibly ... Show Full Article WASHINGTON, Aug. 18 -- The Senate Health, Education, Labor and Pensions Subcommittee on Employment and Workplace Safety released the following written testimony by Jonathan Liebert, CEO of Better Business Bureau of Southern Colorado, Colorado Springs, from a July 29, 2026, hearing entitled "The Impact of AI on the Workforce": * * * Artificial intelligence is already changing how Americans work, learn, make decisions, and compete. The central workforce challenge is no longer whether AI will be adopted; it is whether workers, employers, educators, and communities will be prepared to use it responsiblyand share in its benefits.
"AI is not an Oracle; it is an Ideation Engine"
Since 2025, I have trained over 5,500 people across Colorado and around the country on the practical and responsible use of AI. That work has given me a front-row seat to extraordinary curiosity and real anxiety.
Business owners want to know which tools are credible. Nonprofit leaders want to know how to expand impact without compromising privacy. Educators want to know how to prepare students for jobs that are changing faster than curricula. Workers want to know whether AI will replace them or help them become more capable.
Better Business Bureau of Southern Colorado created the BBB AI Hub (visit it here at www.bbbaihub.org) to help close the gap between the speed of technological change and the ability of people and organizations to respond. In collaboration with the Pikes Peak Workforce Center, the UCCS College of Business, Exponential Impact, the Colorado Springs Chamber & EDC, the Pikes Peak Small Business Development Center, the City of Colorado Springs, the Colorado AI Alliance, the Pikes Peak Business Alliance (PPBEA), and the National Association of Workforce Boards, the Hub connects AI literacy, responsible adoption, governance, and workforce preparation. Our experience suggests that the nation does not need to build an entirely new delivery system. Trusted local infrastructure already exists in workforce centers, community colleges, libraries, BBBs, chambers, and nonprofit networks. It needs practical curriculum, capacity, coordination, and sustained support.
My recommendations are straightforward: invest in rapid AI literacy through trusted local intermediaries; provide practical governance and vendor-evaluation tools for small organizations; improve task-level workforce data; prepare educators and students for AI-enabled work; protect human accountability in consequential employment decisions; and ensure rural communities, small employers, nonprofits, and workers outside technology centers can participate in the opportunity rather than absorb only the disruption.
I. A Cautiously Optimistic View of AI
My position is neither uncritically pro-AI nor reflexively anti-AI. The loudest public voices often occupy the extremes. One side presents AI as an unqualified good that should be adopted as quickly as possible. The other presents it as a force that will inevitably destroy jobs, erode learning, and diminish human value.
Neither perspective is useful to the small-business owner deciding whether to purchase a new platform, the workforce professional helping a displaced worker, the teacher redesigning an assignment, or the nonprofit director with three employees and no technology department.
Cautious optimism means holding two realities at the same time. AI can improve productivity, reduce administrative burden, strengthen customer service, broaden access to expertise, and help small organizations accomplish work that previously required larger teams. I have watched people use AI to organize complex information, improve communication, accelerate research, and turn ideas into usable plans.
I have also watched people enter sensitive customer information into public systems, accept fabricated citations as fact, rely on incomplete or biased outputs, and purchase products marketed as "AI-powered" without understanding what the technology actually did. These are not abstract concerns. They are predictable consequences of adoption moving faster than education and governance.
The question before us is not simply whether AI is good or bad. The better questions are: Who will understand it? Who will have access to useful training? Who will be accountable when it influences a consequential decision? Which communities will receive the productivity gains, and which will bear the disruption? Public policy should help answer those questions before market forces answer them by default.
II. What I Am Seeing in Classrooms and Communities Since 2025, I have trained over 5,500 individuals through classes, workshops, presentations, and community conversations across Colorado and around the country. The audiences have included business owners, employees, educators, nonprofit leaders, workforce professionals, students, and community leaders.
Although their industries and responsibilities differ, the same themes appear repeatedly.
* People are interested in AI, but many do not know where to begin or which information to trust.
* Many organizations are already using AI informally, even when leadership believes adoption has not started.
* Employees frequently receive access to tools before receiving guidance about privacy, verification, intellectual property, bias, or human review.
* Small organizations are overwhelmed by vendor claims and lack the legal, technical, and procurement capacity available to large enterprises.
* Educators recognize that AI will be part of students' futures, but they need support to distinguish responsible learning from shortcut-taking and academic dishonesty.
* Fear often declines when people receive hands-on instruction and a realistic understanding of both capabilities and limitations.
That is the posture I encourage in every room: curiosity about AI produces better questions, better verification, and better decisions than either uncritical enthusiasm or reflexive dismissal.
The most common problem I observe is not that people refuse to use AI. It is that they trust it too quickly or use it without enough context. AI can become a cognitive crutch that replaces thought, or a catalyst that improves thought. The difference is not the tool alone; it is the quality of instruction, the expectations around its use, and the presence of human judgment.
The framing I return to most often in class is this: AI is not an oracle, it is an ideation engine. An oracle is consulted for answers and believed. An ideation engine is used to generate options, surface considerations a person had not thought of, stress-test an assumption, draft a first version, or restate a complex idea in plainer terms. The output is raw material for thinking, not a conclusion to be adopted. When people treat AI as an oracle, they outsource judgment and inherit whatever errors, gaps, or biases the system carried. When they treat it as an ideation engine, they stay in the decision, and the technology makes them faster and more thorough without making them less accountable.
This distinction has practical consequences in the workplace. A manager who asks AI to draft three approaches to a staffing problem, then applies experience and knowledge of the team to choose among them, has used the tool well. A manager who asks AI who to lay off and acts on the answer has not. The same is true of a nurse checking a care protocol, a small-business owner reviewing a contract summary, or a teacher building a lesson plan. In every case the value comes from the human bringing context the system does not have--local knowledge, ethical judgment, and accountability for the outcome.
III. The BBB AI Hub: A Local Model for Workforce Preparation Better Business
Bureau has spent more than a century helping people make informed marketplace decisions and helping businesses build trust. AI is now changing both sides of that mission. It is changing how companies operate, how consumers encounter information, how products are marketed, and how fraud and misinformation can be created and scaled. For BBB of Southern Colorado, AI literacy and AI governance are therefore a natural extension of marketplace trust.
We created the BBB AI Hub to provide practical, accessible, and responsible AI education. The Hub is designed for people who do not need to become engineers but do need to make better decisions about AI. Its work centers on four connected priorities:
* * *
Priority ... What It Means
AI literacy ... Understanding what AI can and cannot do; writing effective instructions; verifying outputs; protecting sensitive information; and recognizing misinformation, bias, and fabricated content.
Workforce preparation ... Helping workers use AI to become more capable, helping employers redesign work thoughtfully, and connecting training to actual roles and advancement opportunities.
AI governance ... Creating clear rules about approved tools, restricted data, human review, accountability, documentation, and employee training.
Marketplace trust ... Helping organizations evaluate AI vendors and claims before making purchasing or implementation decisions.
* * *
Our collaboration with the Pikes Peak Workforce Center is especially important because AI education and workforce services are too often treated as separate activities. They should be connected. Workforce centers understand local employers, job seekers, occupations, and barriers to employment. Community-based organizations understand trust and access. Education providers understand learning. When these capabilities are aligned, AI training can be connected to real jobs, incumbent-worker development, reemployment, and advancement rather than existing as a stand-alone technology demonstration.
This partnership also offers a broader lesson for federal policy: the country already has local institutions capable of reaching workers and small employers. Federal support should strengthen these institutions rather than creating programs that are difficult to access, overly technical, or disconnected from local labor-market needs.
IV. AI Literacy Must Become a Foundational Workforce Skill AI literacy should not be confused with learning one product. Tools will change. Interfaces will change. The durable skill is knowing how to work with AI systems critically and responsibly. At a minimum, workers and employers should understand:
** How to describe a task clearly and provide appropriate context.
* How to evaluate whether AI is suitable for the task at all.
* How to verify factual claims, calculations, citations, and recommendations.
* How to protect customer, employee, student, patient, financial, and proprietary information.
* How bias can enter data, prompts, outputs, and decisions.
* When human review is required and who remains accountable.
* How to document and communicate the use of AI when transparency matters.
AI literacy should be role-based. A teacher, human-resources manager, small-business owner, customerservice employee, and cybersecurity professional do not need the same training. Organizations should identify the tasks people actually perform, the risks attached to those tasks, and the level of review required.
This is more effective than generic training and more realistic for employers with limited time and resources.
AI literacy is also inseparable from critical thinking. The presence of AI does not make critical thinking less important; it makes it more visible. A person must decide whether the question is well framed, whether the output is credible, what information is missing, and whether the result should influence a human decision.
Responsible AI use requires more judgment, not less.
V. AI Governance for Organizations of Every Size Many organizations are waiting for perfect certainty before establishing an AI policy. That is a mistake.
Governance does not require a large compliance department. It begins with practical decisions that any organization can make.
1. Adopt an AI use policy -- Define approved and prohibited uses, identify restricted information, and establish expectations for disclosure and human review.
2. Create a role-based literacy plan -- Train employees according to the decisions they make, the data they handle, and the risks associated with their work.
3. Require accountable human oversight -- AI may support a decision, but a named person should remain responsible when the decision affects employment, safety, education, finances, or access to services.
4. Assess and improve continuously -- Tools, risks, and regulations will change. Organizations should review their practices, incidents, and training on a regular schedule.
The goal of governance is not to prevent experimentation. It is to create a safe environment for experimentation. Clear boundaries give employees confidence about what they may do, protect the organization from avoidable risk, and help leadership learn from use rather than discover it only after a problem occurs.
VI. Vetting AI Businesses and Tools
The rapid growth of AI has created a marketplace challenge. "AI-powered" has no consistent practical meaning for a buyer. Some products use sophisticated models in meaningful ways. Others add a thin AI feature to an existing platform. Still others rely on marketing claims that a small employer cannot independently verify.
Large companies may have procurement teams, legal counsel, cybersecurity experts, data-governance professionals, and the leverage to negotiate contracts. Small employers and nonprofits frequently have none of these. They need plain-language standards and questions that help them evaluate risk before purchasing or connecting a tool to sensitive systems.
Questions every organization should ask an AI vendor
Area ... Questions to Ask
Data practices... What information is collected? Where is it stored? Is customer data used to train the system? Can data be deleted or exported?
Security ... What security controls and independent assessments are in place? How are incidents reported?
Performance ... What does the product actually do, and what evidence supports claims about accuracy, productivity, or cost savings?
Human accountability ... What happens when the system is wrong? Can a human review, override, and explain the result?
Bias and accessibility ... How has the system been evaluated across different users and populations?
Contract terms ... Who owns inputs and outputs? What happens to data when the contract ends? What indemnification or limitations apply?
Vendor credibility ... Who operates the company? Is support available? Are references and claims independently verifiable?
* * *
Trusted marketplace organizations can help translate technical questions into usable guidance. This is not a substitute for regulation, cybersecurity, or legal review. It is a practical layer of consumer and business education that can reduce harm before it occurs.
VII. Preparing Educators and Students for an AI-Enabled Workforce The workforce conversation cannot begin after graduation. Educators are being asked to prepare students for a labor market in which AI will be embedded across occupations, including roles that are not traditionally considered technical. At the same time, schools are navigating legitimate concerns about academic integrity, student privacy, unequal access, and overreliance on automated tools.
The answer is not to pretend AI is absent from the world students will enter. Nor is it to allow unrestricted use without expectations. Schools need age-appropriate AI literacy, clear policies, teacher professional development, and learning experiences that preserve foundational knowledge while teaching students how to use AI as a thought partner rather than a substitute for thought.
The most durable workforce skills will include communication, critical thinking, ethical judgment, creativity, adaptability, domain expertise, and the ability to evaluate AI-generated work. Students should learn how to ask better questions, challenge an output, identify missing context, and explain their own reasoning. These capabilities will help them whether they become nurses, entrepreneurs, technicians, teachers, public servants, or software developers.
VIII. Workforce Disruption, Opportunity, and Fairness Some jobs will be redesigned, and some tasks--particularly routine, entry-level, and administrative tasks-- will be reduced or automated. New roles will emerge, and existing workers may become more productive and capable. But disruption and opportunity are often the same event viewed from different sides.
The honest concern is that new opportunities may not appear in the same communities, at the same wages, or quickly enough for workers whose tasks are displaced. Productivity gains may accrue to organizations without being reflected in job quality, wages, advancement, or reduced workload. Rural communities and smaller employers may face the disruption without having equal access to training, infrastructure, or expertise.
Workers should have a meaningful voice in how AI is introduced into their jobs. Employers should communicate what is changing, what data is being used, how performance will be evaluated, and how employees can challenge an incorrect automated conclusion. Transparency and human accountability are especially important when AI materially affects hiring, promotion, scheduling, performance evaluation, discipline, layoffs, or workplace surveillance.
The message I bring to every class is practical: learn to use these tools strategically, ethically, and responsibly, and they can become leverage. They can help a worker perform the current job more effectively, preserve time for higher-value work, or compete for opportunities that were previously out of reach. That outcome is possible, but it is not automatic. Preparation determines who benefits.
IX. Recommendations to Congress
1. Fund rapid, practical AI upskilling -- Use trusted local intermediaries such as workforce centers, community colleges, libraries, BBBs, chambers, and nonprofit networks to deliver short-cycle, role-based training tied to employment and advancement outcomes.
2. Strengthen AI literacy for educators and students -- Support teacher professional development, age-appropriate curricula, privacy protections, and career-connected learning that prepares students to work responsibly with AI.
3. Equip small organizations with governance tools -- Develop and distribute model AI-use policies, vendor-evaluation guides, cybersecurity practices, and human-oversight frameworks that small businesses and nonprofits can realistically adopt.
4. Improve workforce intelligence -- Create timely, task-level, and geographically detailed data so policymakers and local workforce systems can identify disruption before layoffs and align training with changing employer demand.
5. Preserve transparency and human accountability -- When AI materially affects hiring, evaluation, discipline, layoffs, scheduling, or surveillance, workers should know it is being used and a person should remain accountable for the decision.
6. Ensure broad access to the benefits -- Prioritize rural communities, small employers, underserved populations, and workers outside major technology centers so they receive training and opportunity, not only disruption.
7. Evaluate outcomes, not just adoption -- Measure job quality, wage mobility, worker advancement, productivity, small-business competitiveness, safety, and access--not merely how many organizations purchased an AI tool.
Conclusion
AI itself is not the greatest workforce threat. The greater risk is allowing the technology to advance while workers, businesses, nonprofits, educators, and communities are left without the knowledge, protections, or support required to use it well.
America does not need to choose between innovation and responsibility. We can pursue both. We can help businesses compete while protecting workers. We can prepare students for AI-enabled careers while preserving the importance of human learning. We can encourage experimentation while requiring accountability. We can be optimistic about what AI makes possible and serious about the conditions required for those possibilities to benefit people broadly.
The future of work will not be predetermined by an algorithm. It will be shaped by the choices we make about who is educated, who is protected, who has access, who remains accountable, and who shares in the value that AI helps create.
Chairman Banks, Ranking Member Hickenlooper, and members of the Subcommittee, thank you for the opportunity to submit this testimony and for your attention to the workers, employers, educators, and communities navigating this transition.
* * *
Original text here: https://www.help.senate.gov/imo/media/doc/6cbbd241-b14f-d7fb-a82f-91145ad11dc1/Liebert%20Testimony_c03b7053-d741-46c3-a80a-442da4ed5114.pdf