Think Tanks
Here's a look at documents from think tanks
Think Tanks
Featured Stories
Center of the American Experiment Issues Commentary: Investigative Journalists Appear on Charges Related to Exposing Vulnerabilities in Minnesota's Voter Vouching System
MINNETONKA, Minnesota, Nov. 18 -- The Center of the American Experiment, a civic and educational organization that says it creates and advocates policies, issued the following commentary:
* * *
Investigative journalists appear on charges related to exposing vulnerabilities in Minnesota's voter vouching system
Written by David Zimmer
November 16, 2026
Investigative journalists James O'Keefe and Cameron Higby made their first appearances in Hennepin County District Court this afternoon, in response to criminal charges and summonses issued by Hennepin County Attorney Mary Moriarty in October. ... Show Full Article MINNETONKA, Minnesota, Nov. 18 -- The Center of the American Experiment, a civic and educational organization that says it creates and advocates policies, issued the following commentary: * * * Investigative journalists appear on charges related to exposing vulnerabilities in Minnesota's voter vouching system Written by David Zimmer November 16, 2026 Investigative journalists James O'Keefe and Cameron Higby made their first appearances in Hennepin County District Court this afternoon, in response to criminal charges and summonses issued by Hennepin County Attorney Mary Moriarty in October.The charges include allegations of identity theft and unlawful voting - the result of O'Keefe and Higby's undercover video exposing vulnerabilities in Minnesota's vouching system during the primary election in August. Judge Luis Bartolomei presided over the first appearance calendar, including these cases.
O'Keefe, who lives in Florida, was represented by attorney Chris Madel, while Higby who lives in Washington state was represented by attorney Ryan Gary.
Morgan Kunz, the Hennepin County Attorney's Office (HCAO) Criminal Division Director appeared on behalf of the HCAO - an indication that these charges are not standard in any way.
Madel, who has successfully spared with Moriarty on other cases (most notably the now dismissed murder case against on-duty Minnesota State Trooper Ryan Londregan), has argued that the charges against O'Keefe (and Higby) are politically motivated and inappropriate.
* * *
"These charges are Mary Moriarty's disgraceful attempt to turn her politics into a criminal conviction."
Attorney Chris Madel
* * *
During today's hearing, Kunz requested that three conditions be placed on the defendants:
1) No contact with any of the alleged "victims" in the case (presumably people whose names were referred to by O'Keefe and Higby during the undercover act.
2) No contact with any Minnesota elections judges
3) That the defendants have no contact with each other.
Madel and Gary stipulated to the first two requests but successfully argued that the HCAO was alleging a conspiracy had occurred between the defendants and that it would be essential that they be able to communicate as they assemble their defense. Judge Bartolomei maintained the personal recognizance release status on both defendants and declined the request for a no contact order between the defendants at this time.
The next court appearance for both defendants is scheduled for January 13th at 9 am for an omnibus hearing where probable cause and the admissibility of evidence will be challenged.
* * *
David Zimmer is a Public Safety Policy Fellow at Center of the American Experiment.
* * *
Original text here: https://www.americanexperiment.org/investigative-journalists-appear-on-charges-related-to-exposing-vulnerabilities-in-minnesotas-voter-vouching-system/
[Category: ThinkTank]
* * *
Investigative journalists appear on charges related to exposing vulnerabilities in Minnesota's voter vouching system
Written by David Zimmer
November 16, 2026
Investigative journalists James O'Keefe and Cameron Higby made their first appearances in Hennepin County District Court this afternoon, in response to criminal charges and summonses issued by Hennepin County Attorney Mary Moriarty in October. ... Show Full Article MINNETONKA, Minnesota, Nov. 18 -- The Center of the American Experiment, a civic and educational organization that says it creates and advocates policies, issued the following commentary: * * * Investigative journalists appear on charges related to exposing vulnerabilities in Minnesota's voter vouching system Written by David Zimmer November 16, 2026 Investigative journalists James O'Keefe and Cameron Higby made their first appearances in Hennepin County District Court this afternoon, in response to criminal charges and summonses issued by Hennepin County Attorney Mary Moriarty in October.The charges include allegations of identity theft and unlawful voting - the result of O'Keefe and Higby's undercover video exposing vulnerabilities in Minnesota's vouching system during the primary election in August. Judge Luis Bartolomei presided over the first appearance calendar, including these cases.
O'Keefe, who lives in Florida, was represented by attorney Chris Madel, while Higby who lives in Washington state was represented by attorney Ryan Gary.
Morgan Kunz, the Hennepin County Attorney's Office (HCAO) Criminal Division Director appeared on behalf of the HCAO - an indication that these charges are not standard in any way.
Madel, who has successfully spared with Moriarty on other cases (most notably the now dismissed murder case against on-duty Minnesota State Trooper Ryan Londregan), has argued that the charges against O'Keefe (and Higby) are politically motivated and inappropriate.
* * *
"These charges are Mary Moriarty's disgraceful attempt to turn her politics into a criminal conviction."
Attorney Chris Madel
* * *
During today's hearing, Kunz requested that three conditions be placed on the defendants:
1) No contact with any of the alleged "victims" in the case (presumably people whose names were referred to by O'Keefe and Higby during the undercover act.
2) No contact with any Minnesota elections judges
3) That the defendants have no contact with each other.
Madel and Gary stipulated to the first two requests but successfully argued that the HCAO was alleging a conspiracy had occurred between the defendants and that it would be essential that they be able to communicate as they assemble their defense. Judge Bartolomei maintained the personal recognizance release status on both defendants and declined the request for a no contact order between the defendants at this time.
The next court appearance for both defendants is scheduled for January 13th at 9 am for an omnibus hearing where probable cause and the admissibility of evidence will be challenged.
* * *
David Zimmer is a Public Safety Policy Fellow at Center of the American Experiment.
* * *
Original text here: https://www.americanexperiment.org/investigative-journalists-appear-on-charges-related-to-exposing-vulnerabilities-in-minnesotas-voter-vouching-system/
[Category: ThinkTank]
Center of the American Experiment Issues Commentary: Controversial Ethnic Studies License Will Get a Public Hearing
MINNETONKA, Minnesota, Nov. 18 -- The Center of the American Experiment, a civic and educational organization that says it creates and advocates policies, issued the following commentary:
* * *
Controversial Ethnic Studies license will get a public hearing
Written by Catrin Wigfall
November 17, 2026
Thank you to those who reached out to the Professional Educator Licensing and Standards Board (PELSB) and requested a hearing on proposed licensing rules for teaching ethnic studies in Minnesota public schools.
PELSB received around 275 comments on the proposed ethnic studies licensure, the vast ... Show Full Article MINNETONKA, Minnesota, Nov. 18 -- The Center of the American Experiment, a civic and educational organization that says it creates and advocates policies, issued the following commentary: * * * Controversial Ethnic Studies license will get a public hearing Written by Catrin Wigfall November 17, 2026 Thank you to those who reached out to the Professional Educator Licensing and Standards Board (PELSB) and requested a hearing on proposed licensing rules for teaching ethnic studies in Minnesota public schools. PELSB received around 275 comments on the proposed ethnic studies licensure, the vastmajority of them in opposition. The board has confirmed that a virtual hearing will be conducted by Administrative Law Judge (ALJ) Kristien Butler on Jan. 8, 2027 beginning at 9:30 a.m. The hearing will continue until 4:30 p.m. or until all those who wish to testify have been heard, whichever occurs first.
The subject matter standards in the proposed licensing rules are, according to PELSB, "based on disciplinary concepts taught at universities across the country" and do not include "references to any specific belief systems, political ideologies, or geopolitical events." "Furthermore, the standards are merely a set of knowledge teachers need for licensure."
I would argue that characterization doesn't fully hold up. The standards require teachers to "understand and apply critical frameworks" and analyze through "decolonial and postcolonial lenses." That is not framework-neutral. Other standards treat contested claims as settled content, such as expecting teachers to understand counternarratives that "deconstruct dominant narratives" and to connect "educational debt" to the "erasure of ethnic studies in Minnesota." There are also no definitions of contested terms such as "white supremacy," "settler-colonialism," "neocolonialism," "oppression," and "decolonial."
PELSB continued:
Rather, the standards reflect a baseline of understanding teachers of ethnic studies believe their colleagues need for licensure. Standard A (8) also states teachers must "understand the history and plurality of the disciplines and pedagogies of ethnic studies," versus one specific way of teaching ethnic studies.
The only illustrations Standard A (8) offers are one historical lineage (the Third World Liberation Front) and one distinction (ethnic studies versus multiculturalism and DEI).
The standards don't dictate what a teacher ends up choosing to do in the classroom, but they do shape how teachers are expected to approach it. The standards' lack of value neutrality presents a real dilemma for not only teacher candidates but also teacher preparation providers that will be expected to align their programs to this content.
How to participate in the hearing
The January 8 hearing can be joined via this webinar link. If an access code is requested, it is: 2498 396 3237. The webinar password is: D2FhSnmPR98. If you plan to dial in via a phone or video system, the password is: 32347667.
Attendees will be able to share oral comments during the hearing. Those wishing to submit written comments can do so through the Minnesota Court of Administrative Hearings' e-comments system. According to PELSB, the public will be informed when the comment portal is available and will be provided more information on how to share public comments and statements before the ALJ.
American Experiment will participate both orally and in writing, focusing primarily on these concerns:
* The proposed standards extend well beyond the Minnesota Legislature's statutory definition of ethnic studies.
* The Statement of Need and Reasonableness (SONAR) does not affirmatively demonstrate the need for and reasonableness of the proposed licensing rules.
* The proposed standards are not neutral content knowledge.
* * *
Catrin Wigfall is a Policy Fellow at Center of the American Experiment.
* * *
Original text here: https://www.americanexperiment.org/controversial-ethnic-studies-license-will-get-a-public-hearing/
[Category: ThinkTank]
* * *
Controversial Ethnic Studies license will get a public hearing
Written by Catrin Wigfall
November 17, 2026
Thank you to those who reached out to the Professional Educator Licensing and Standards Board (PELSB) and requested a hearing on proposed licensing rules for teaching ethnic studies in Minnesota public schools.
PELSB received around 275 comments on the proposed ethnic studies licensure, the vast ... Show Full Article MINNETONKA, Minnesota, Nov. 18 -- The Center of the American Experiment, a civic and educational organization that says it creates and advocates policies, issued the following commentary: * * * Controversial Ethnic Studies license will get a public hearing Written by Catrin Wigfall November 17, 2026 Thank you to those who reached out to the Professional Educator Licensing and Standards Board (PELSB) and requested a hearing on proposed licensing rules for teaching ethnic studies in Minnesota public schools. PELSB received around 275 comments on the proposed ethnic studies licensure, the vastmajority of them in opposition. The board has confirmed that a virtual hearing will be conducted by Administrative Law Judge (ALJ) Kristien Butler on Jan. 8, 2027 beginning at 9:30 a.m. The hearing will continue until 4:30 p.m. or until all those who wish to testify have been heard, whichever occurs first.
The subject matter standards in the proposed licensing rules are, according to PELSB, "based on disciplinary concepts taught at universities across the country" and do not include "references to any specific belief systems, political ideologies, or geopolitical events." "Furthermore, the standards are merely a set of knowledge teachers need for licensure."
I would argue that characterization doesn't fully hold up. The standards require teachers to "understand and apply critical frameworks" and analyze through "decolonial and postcolonial lenses." That is not framework-neutral. Other standards treat contested claims as settled content, such as expecting teachers to understand counternarratives that "deconstruct dominant narratives" and to connect "educational debt" to the "erasure of ethnic studies in Minnesota." There are also no definitions of contested terms such as "white supremacy," "settler-colonialism," "neocolonialism," "oppression," and "decolonial."
PELSB continued:
Rather, the standards reflect a baseline of understanding teachers of ethnic studies believe their colleagues need for licensure. Standard A (8) also states teachers must "understand the history and plurality of the disciplines and pedagogies of ethnic studies," versus one specific way of teaching ethnic studies.
The only illustrations Standard A (8) offers are one historical lineage (the Third World Liberation Front) and one distinction (ethnic studies versus multiculturalism and DEI).
The standards don't dictate what a teacher ends up choosing to do in the classroom, but they do shape how teachers are expected to approach it. The standards' lack of value neutrality presents a real dilemma for not only teacher candidates but also teacher preparation providers that will be expected to align their programs to this content.
How to participate in the hearing
The January 8 hearing can be joined via this webinar link. If an access code is requested, it is: 2498 396 3237. The webinar password is: D2FhSnmPR98. If you plan to dial in via a phone or video system, the password is: 32347667.
Attendees will be able to share oral comments during the hearing. Those wishing to submit written comments can do so through the Minnesota Court of Administrative Hearings' e-comments system. According to PELSB, the public will be informed when the comment portal is available and will be provided more information on how to share public comments and statements before the ALJ.
American Experiment will participate both orally and in writing, focusing primarily on these concerns:
* The proposed standards extend well beyond the Minnesota Legislature's statutory definition of ethnic studies.
* The Statement of Need and Reasonableness (SONAR) does not affirmatively demonstrate the need for and reasonableness of the proposed licensing rules.
* The proposed standards are not neutral content knowledge.
* * *
Catrin Wigfall is a Policy Fellow at Center of the American Experiment.
* * *
Original text here: https://www.americanexperiment.org/controversial-ethnic-studies-license-will-get-a-public-hearing/
[Category: ThinkTank]
Center of the American Experiment Issues Commentary: Companies Aren't Taxpayers, They are Tax Collectors
MINNETONKA, Minnesota, Nov. 18 -- The Center of the American Experiment, a civic and educational organization that says it creates and advocates policies, issued the following commentary:
* * *
Companies aren't taxpayers, they are tax collectors
Written by John Phelan
November 16, 2026
"I have been clear from the very beginning that I'm not going to increase taxes given where Minnesota sits vis-a-vis the rest of the country," Governor-elect Amy Klobuchar said in an interview recorded for "At Issue with Tom Hauser" shortly after her election. "I'm not going to do that."
This may set up a struggle ... Show Full Article MINNETONKA, Minnesota, Nov. 18 -- The Center of the American Experiment, a civic and educational organization that says it creates and advocates policies, issued the following commentary: * * * Companies aren't taxpayers, they are tax collectors Written by John Phelan November 16, 2026 "I have been clear from the very beginning that I'm not going to increase taxes given where Minnesota sits vis-a-vis the rest of the country," Governor-elect Amy Klobuchar said in an interview recorded for "At Issue with Tom Hauser" shortly after her election. "I'm not going to do that." This may set up a strugglewith her DFL colleagues in the state legislature who ran on commitments of new spending and taxes totaling $14 billion each year.
I noted last week that House Leader Zack Stephenson is already pushing for a new tax on social media use in Minnesota. DFL Senate Majority Leader Erin Murphy has joined the push. "The Senate, of course, has proposed a social media tax, which is not a tax on people but on the data of social media companies," she told WCCO. "We're interested in that."
Sen. Murphy is wrong. While a government can impose a tax obligation on a company, it cannot control who actually bears the burden of that tax; the "incidence." Nobody, for example, believes that cigarette companies bear the burden of the taxes imposed on their sales of their product; that burden is passed on to the consumer and becomes a tax on their consumption of the product.
As one of my old textbooks, "Economics of the Public Sector" by Joseph E. Stiglitz puts it:
...economic reality--for better or worse--does not always follow the laws passed by legislatures. Thus, economists distinguish between those who bear the burden of a tax and those on whom a tax is imposed or levied. The tax burden is the true economic weight of a tax...Economists use a more neutral word to describe the effects of taxation--they ask, what is the incidence of the tax?
Tech companies like Meta, which generate most of their revenues from targeted advertising, have several ways they could pass the burden of Minnesota's social media tax on to their customers.
They could raise the cost of ad placements, bidding minimums, or sponsored posts for businesses and agencies targeting users in Minnesota; add a state-specific regulatory fee or tax surcharge directly onto advertising invoices for campaigns running within our state; incentivize or accelerate paid, ad-free subscription tiers to shift the direct financial cost of the service away from advertisers and onto individual end-users; and adjust their payout formulas or revenue-sharing percentages for content creators, publishers, or developers, lower for traffic originating from our state.
In all cases, the tech company would pass the burden on to its customers, who would attempt to pass it onto their customers, Minnesota's residents.
A tax on Minnesotans social media use would be a breach of Gov.-elect Klobuchar's "no new taxes" pledge. Sen. Murphy cannot escape the simple truth that companies are not taxpayers, they are tax collectors, and they collect their taxes from you.
* * *
John Phelan is an Economist at the Center of the American Experiment.
* * *
Original text here: https://www.americanexperiment.org/companies-arent-taxpayers-they-are-tax-collectors/
[Category: ThinkTank]
* * *
Companies aren't taxpayers, they are tax collectors
Written by John Phelan
November 16, 2026
"I have been clear from the very beginning that I'm not going to increase taxes given where Minnesota sits vis-a-vis the rest of the country," Governor-elect Amy Klobuchar said in an interview recorded for "At Issue with Tom Hauser" shortly after her election. "I'm not going to do that."
This may set up a struggle ... Show Full Article MINNETONKA, Minnesota, Nov. 18 -- The Center of the American Experiment, a civic and educational organization that says it creates and advocates policies, issued the following commentary: * * * Companies aren't taxpayers, they are tax collectors Written by John Phelan November 16, 2026 "I have been clear from the very beginning that I'm not going to increase taxes given where Minnesota sits vis-a-vis the rest of the country," Governor-elect Amy Klobuchar said in an interview recorded for "At Issue with Tom Hauser" shortly after her election. "I'm not going to do that." This may set up a strugglewith her DFL colleagues in the state legislature who ran on commitments of new spending and taxes totaling $14 billion each year.
I noted last week that House Leader Zack Stephenson is already pushing for a new tax on social media use in Minnesota. DFL Senate Majority Leader Erin Murphy has joined the push. "The Senate, of course, has proposed a social media tax, which is not a tax on people but on the data of social media companies," she told WCCO. "We're interested in that."
Sen. Murphy is wrong. While a government can impose a tax obligation on a company, it cannot control who actually bears the burden of that tax; the "incidence." Nobody, for example, believes that cigarette companies bear the burden of the taxes imposed on their sales of their product; that burden is passed on to the consumer and becomes a tax on their consumption of the product.
As one of my old textbooks, "Economics of the Public Sector" by Joseph E. Stiglitz puts it:
...economic reality--for better or worse--does not always follow the laws passed by legislatures. Thus, economists distinguish between those who bear the burden of a tax and those on whom a tax is imposed or levied. The tax burden is the true economic weight of a tax...Economists use a more neutral word to describe the effects of taxation--they ask, what is the incidence of the tax?
Tech companies like Meta, which generate most of their revenues from targeted advertising, have several ways they could pass the burden of Minnesota's social media tax on to their customers.
They could raise the cost of ad placements, bidding minimums, or sponsored posts for businesses and agencies targeting users in Minnesota; add a state-specific regulatory fee or tax surcharge directly onto advertising invoices for campaigns running within our state; incentivize or accelerate paid, ad-free subscription tiers to shift the direct financial cost of the service away from advertisers and onto individual end-users; and adjust their payout formulas or revenue-sharing percentages for content creators, publishers, or developers, lower for traffic originating from our state.
In all cases, the tech company would pass the burden on to its customers, who would attempt to pass it onto their customers, Minnesota's residents.
A tax on Minnesotans social media use would be a breach of Gov.-elect Klobuchar's "no new taxes" pledge. Sen. Murphy cannot escape the simple truth that companies are not taxpayers, they are tax collectors, and they collect their taxes from you.
* * *
John Phelan is an Economist at the Center of the American Experiment.
* * *
Original text here: https://www.americanexperiment.org/companies-arent-taxpayers-they-are-tax-collectors/
[Category: ThinkTank]
CSIS Issues Critical Questions Q&A: What AI Agent Attacks Mean for India's Cyberspace
WASHINGTON, Nov. 18 -- The Center for Strategic and International Studies issued the following Critical Questions Q&A on Nov. 16, 2026, involving fellow Aman Thakker and program coordinator and research assistant Aditya Goswamy of the Chair on India and Emerging Asia Economics:
* * *
What AI Agent Attacks Mean for India's Cyberspace
On October 15, India disclosed its first cyber incident where an AI agent, not a human operator, was responsible. The Government of India issued new guidance on November 4 for auditing AI systems and reporting future incidents. With frontier AI development concentrated ... Show Full Article WASHINGTON, Nov. 18 -- The Center for Strategic and International Studies issued the following Critical Questions Q&A on Nov. 16, 2026, involving fellow Aman Thakker and program coordinator and research assistant Aditya Goswamy of the Chair on India and Emerging Asia Economics: * * * What AI Agent Attacks Mean for India's Cyberspace On October 15, India disclosed its first cyber incident where an AI agent, not a human operator, was responsible. The Government of India issued new guidance on November 4 for auditing AI systems and reporting future incidents. With frontier AI development concentratedin the United States and AI diffusion running through India and the Global South, New Delhi and Washington need to coordinate more in this area. As Prime Minister Modi and President Trump prepare to meet twice before year-end, there is a window for both countries to develop a blueprint for reporting, governing, and containing future incidents.
Q1: What happened?
A1: In its October 15 "critical" severity advisory, India's Computer Emergency Response Team (CERT-In) identified two distinct types of AI agent incidents against Indian technology assets. The first involved AI agents developed by frontier labs that exceeded the scope of their assigned evaluation tasks during model research and development, escaped containment, and accessed live Indian systems, similar to the mid-2026 OpenAI-Hugging-Face incident. It is a serious but tractable threat, but is addressable through stronger sandbox design and authorization controls. The second type of attack is more consequential: criminal actors deliberately weaponizing open-source agentic AI tools to conduct "reconnaissance, exploitation, and financially motivated intrusions," compressing attack timelines from days to hours.
Q2: Why does this matter?
A2: In every incident CERT-In documented, the breach came to light either through voluntary disclosures from AI providers or through CERT-In's own threat hunting. None of the affected organizations detected the intrusions, even as agents accessed and read personal data and confidential business records.
Legacy network infrastructures, fragmented security controls, limited automation, and a talent gap mean only 7 percent of Indian organizations have achieved a "mature" cybersecurity readiness rating. AI is compounding these gaps: 68 percent of Indian organizations have limited or no AI and automation in security operations, and those without it take an average of 311 days to identify and contain a breach, 55 more than those with extensive AI deployment.
India is racing toward a $1 trillion digital economy, and sees AI as a key accelerant. Sustaining that growth will require investor confidence in India's ability to govern AI risks as adeptly and quickly as it adopts AI.
Q3: How has the Indian government responded?
A3: The Indian government has been proactively engaged on the threat of AI-enabled cyber. CERT-In's Comprehensive Cyber Security Audit Policy Guidelines, published in July 2025, established a structured audit framework for critical infrastructure; in April 2026, CERT-In issued a high-severity advisory (CIAD-2026-0020) flagging autonomous reconnaissance and multistage attack orchestration as systemic threats. Finance Minister Nirmala Sitharaman convened the country's private and public banking executives to level-set on advancing cybersecurity standards after news of the cyber capabilities of Anthropic's Mythos class of models became public in May 2026. In July 2026, the Reserve Bank of India issued a directive asking banks and nonbanking financial entities to submit a board-approved assessment of their cyber threat landscape. The October 15 CERT-In disclosure has now accelerated these efforts.
CERT-In's November 4 AI Security Auditing Guidelines represent India's most comprehensive response yet. They call for all AI systems, including those operated by foreign developers, to be classified into four risk tiers: low, moderate, high, and critical. Low-risk systems can self-assess, while moderate- and high-risk systems face annual independent audits. The guidelines also outline requirements for high- and critical-risk systems to face a combination of emergency kill switches, continuous monitoring, and red-team exercises, depending on their tier rankings. For agentic AI systems, they call for all systems to remain within "approved operating parameters" to address sandbox failures. CERT-In also extends its mandate to report all cybersecurity incidents within six hours of any AI-related incidents and demands that enterprises maintain system logs for 180 days.
Q4: How can the United States and India work together to address AI-enabled cyber vulnerabilities?
A4: The labs building the most capable AI systems are overwhelmingly American, but as Indian users and enterprises engage with these systems as early adopters at scale, they will be the ones increasingly absorbing their risks. That gap creates a shared obligation in the U.S.-India relationship. The Quad Leaders' Summit in Manila and the G20 Leaders' Summit in Miami in December give President Trump and Prime Minister Modi two near-term opportunities to advance cooperation on reporting, governing, and containing AI-enabled cyber incidents moving forward.
* Reporting: When Presidents Trump and Xi met in September, they agreed to a bilateral U.S.-China AI incident communication channel. If Washington is building that with Beijing, it should build a more substantive equivalent, with defined thresholds and operating protocols, with allies and partners, including New Delhi.
* Governing: The U.S. Center for Advancing Innovation and Standards for Super Intelligence (CAISSI) is developing AI agent standards and conducting pre-deployment cybersecurity evaluations with frontier labs. CERT-In's November 4 guidelines show India's desire to address audit domains and risk classification, but do not directly address these areas. Establishing mechanisms to align U.S. standards with India's audit framework should be a near-term deliverable for U.S. and Indian policymakers.
* Containing: Both governments bring a myriad of agencies to responding to agentic AI-enabled cyber incidents: for the United States, the National Security Council, CAISSI, and the Cybersecurity and Infrastructure Security Agency; for India, the National Security Council Secretariat, the nascent but not yet fully operational AI Safety Institute, CERT-In, and the Indian Cybercrime Coordination Centre. With technology as new as agentic AI, the rules are being written in real time, with limited shared visibility into how those agencies interact within each country and across borders. Mapping that architecture, developing shared standard operating procedures for cross-border incidents with designated roles for each agency, and scheduling joint tabletop exercises and wargaming scenarios between these agencies should be a near-term deliverable that the National Security Advisor-led Transforming the Relationship Utilizing Strategic Technology initiative should seek to deliver.
* * *
Aman Thakker is a fellow in the Chair on India and Emerging Asia Economics at the Center for Strategic and International Studies (CSIS) in Washington, D.C. Aditya Goswamy a program coordinator and research assistant in the Chair on India and Emerging Asia Economics at CSIS.
* * *
Original text here: https://www.csis.org/analysis/what-ai-agent-attacks-mean-indias-cyberspace
[Category: ThinkTank]
* * *
What AI Agent Attacks Mean for India's Cyberspace
On October 15, India disclosed its first cyber incident where an AI agent, not a human operator, was responsible. The Government of India issued new guidance on November 4 for auditing AI systems and reporting future incidents. With frontier AI development concentrated ... Show Full Article WASHINGTON, Nov. 18 -- The Center for Strategic and International Studies issued the following Critical Questions Q&A on Nov. 16, 2026, involving fellow Aman Thakker and program coordinator and research assistant Aditya Goswamy of the Chair on India and Emerging Asia Economics: * * * What AI Agent Attacks Mean for India's Cyberspace On October 15, India disclosed its first cyber incident where an AI agent, not a human operator, was responsible. The Government of India issued new guidance on November 4 for auditing AI systems and reporting future incidents. With frontier AI development concentratedin the United States and AI diffusion running through India and the Global South, New Delhi and Washington need to coordinate more in this area. As Prime Minister Modi and President Trump prepare to meet twice before year-end, there is a window for both countries to develop a blueprint for reporting, governing, and containing future incidents.
Q1: What happened?
A1: In its October 15 "critical" severity advisory, India's Computer Emergency Response Team (CERT-In) identified two distinct types of AI agent incidents against Indian technology assets. The first involved AI agents developed by frontier labs that exceeded the scope of their assigned evaluation tasks during model research and development, escaped containment, and accessed live Indian systems, similar to the mid-2026 OpenAI-Hugging-Face incident. It is a serious but tractable threat, but is addressable through stronger sandbox design and authorization controls. The second type of attack is more consequential: criminal actors deliberately weaponizing open-source agentic AI tools to conduct "reconnaissance, exploitation, and financially motivated intrusions," compressing attack timelines from days to hours.
Q2: Why does this matter?
A2: In every incident CERT-In documented, the breach came to light either through voluntary disclosures from AI providers or through CERT-In's own threat hunting. None of the affected organizations detected the intrusions, even as agents accessed and read personal data and confidential business records.
Legacy network infrastructures, fragmented security controls, limited automation, and a talent gap mean only 7 percent of Indian organizations have achieved a "mature" cybersecurity readiness rating. AI is compounding these gaps: 68 percent of Indian organizations have limited or no AI and automation in security operations, and those without it take an average of 311 days to identify and contain a breach, 55 more than those with extensive AI deployment.
India is racing toward a $1 trillion digital economy, and sees AI as a key accelerant. Sustaining that growth will require investor confidence in India's ability to govern AI risks as adeptly and quickly as it adopts AI.
Q3: How has the Indian government responded?
A3: The Indian government has been proactively engaged on the threat of AI-enabled cyber. CERT-In's Comprehensive Cyber Security Audit Policy Guidelines, published in July 2025, established a structured audit framework for critical infrastructure; in April 2026, CERT-In issued a high-severity advisory (CIAD-2026-0020) flagging autonomous reconnaissance and multistage attack orchestration as systemic threats. Finance Minister Nirmala Sitharaman convened the country's private and public banking executives to level-set on advancing cybersecurity standards after news of the cyber capabilities of Anthropic's Mythos class of models became public in May 2026. In July 2026, the Reserve Bank of India issued a directive asking banks and nonbanking financial entities to submit a board-approved assessment of their cyber threat landscape. The October 15 CERT-In disclosure has now accelerated these efforts.
CERT-In's November 4 AI Security Auditing Guidelines represent India's most comprehensive response yet. They call for all AI systems, including those operated by foreign developers, to be classified into four risk tiers: low, moderate, high, and critical. Low-risk systems can self-assess, while moderate- and high-risk systems face annual independent audits. The guidelines also outline requirements for high- and critical-risk systems to face a combination of emergency kill switches, continuous monitoring, and red-team exercises, depending on their tier rankings. For agentic AI systems, they call for all systems to remain within "approved operating parameters" to address sandbox failures. CERT-In also extends its mandate to report all cybersecurity incidents within six hours of any AI-related incidents and demands that enterprises maintain system logs for 180 days.
Q4: How can the United States and India work together to address AI-enabled cyber vulnerabilities?
A4: The labs building the most capable AI systems are overwhelmingly American, but as Indian users and enterprises engage with these systems as early adopters at scale, they will be the ones increasingly absorbing their risks. That gap creates a shared obligation in the U.S.-India relationship. The Quad Leaders' Summit in Manila and the G20 Leaders' Summit in Miami in December give President Trump and Prime Minister Modi two near-term opportunities to advance cooperation on reporting, governing, and containing AI-enabled cyber incidents moving forward.
* Reporting: When Presidents Trump and Xi met in September, they agreed to a bilateral U.S.-China AI incident communication channel. If Washington is building that with Beijing, it should build a more substantive equivalent, with defined thresholds and operating protocols, with allies and partners, including New Delhi.
* Governing: The U.S. Center for Advancing Innovation and Standards for Super Intelligence (CAISSI) is developing AI agent standards and conducting pre-deployment cybersecurity evaluations with frontier labs. CERT-In's November 4 guidelines show India's desire to address audit domains and risk classification, but do not directly address these areas. Establishing mechanisms to align U.S. standards with India's audit framework should be a near-term deliverable for U.S. and Indian policymakers.
* Containing: Both governments bring a myriad of agencies to responding to agentic AI-enabled cyber incidents: for the United States, the National Security Council, CAISSI, and the Cybersecurity and Infrastructure Security Agency; for India, the National Security Council Secretariat, the nascent but not yet fully operational AI Safety Institute, CERT-In, and the Indian Cybercrime Coordination Centre. With technology as new as agentic AI, the rules are being written in real time, with limited shared visibility into how those agencies interact within each country and across borders. Mapping that architecture, developing shared standard operating procedures for cross-border incidents with designated roles for each agency, and scheduling joint tabletop exercises and wargaming scenarios between these agencies should be a near-term deliverable that the National Security Advisor-led Transforming the Relationship Utilizing Strategic Technology initiative should seek to deliver.
* * *
Aman Thakker is a fellow in the Chair on India and Emerging Asia Economics at the Center for Strategic and International Studies (CSIS) in Washington, D.C. Aditya Goswamy a program coordinator and research assistant in the Chair on India and Emerging Asia Economics at CSIS.
* * *
Original text here: https://www.csis.org/analysis/what-ai-agent-attacks-mean-indias-cyberspace
[Category: ThinkTank]
CSIS Issues Critical Questions Q&A: Data Is Physical - Its Security Must Be, Too
WASHINGTON, Nov. 18 -- The Center for Strategic and International Studies issued the following Critical Questions Q&A on Nov. 16, 2026, involving A'ndre Gonawela and Taylar Rajic, fellows with the Strategic Technologies Program:
* * *
Data Is Physical; Its Security Must Be, Too
Opening your apartment door with a fob, paying for your groceries with your phone, ordering an Uber, and scrolling through Instagram don't just exist online; they depend on physical infrastructure. Data centers are the invisible backbone of our highly digitalized daily lives. Apps, websites, and AI tools run on physical ... Show Full Article WASHINGTON, Nov. 18 -- The Center for Strategic and International Studies issued the following Critical Questions Q&A on Nov. 16, 2026, involving A'ndre Gonawela and Taylar Rajic, fellows with the Strategic Technologies Program: * * * Data Is Physical; Its Security Must Be, Too Opening your apartment door with a fob, paying for your groceries with your phone, ordering an Uber, and scrolling through Instagram don't just exist online; they depend on physical infrastructure. Data centers are the invisible backbone of our highly digitalized daily lives. Apps, websites, and AI tools run on physicalservers, and data centers house them, store data, and process the computing tasks that keep those digital services running. If a cluster of data centers were destroyed in Northern Virginia, daily life would essentially be on pause: e-commerce, emails, internet, and essentially any app-based service would not be able to function.
Data centers are increasingly becoming targets in conflict precisely because of their strategic significance. Strikes on data centers have been ongoing in geopolitical conflict in the Middle East and Europe, heavily disrupting essential services during wartime. Although the United States is unlikely to experience the same physical threats as Europe or the Middle East, domestic actors have a different but potentially destructive toolkit at their disposal. The physical threats that are plaguing current geopolitical conflicts--missile strikes and kinetic attacks in active war--are not a concern to U.S.-based data centers. Insider threats and coordinated attacks on facilities, conducted by both domestic actors and foreign-enabled proxies, pose the biggest risks, and are currently severely underappreciated.
Q1: What happened in the attacks on Russia's Yandex data centers? Have we seen these types of strikes before?
A1: From October 8-11, 2026, Ukrainian drones struck three data centers operated by Russian technology company Yandex as part of the ongoing Russia-Ukraine war. Although no deaths were reported, the strikes damaged three of the five data centers operated by the Russian tech giant, seriously disrupting daily life in Russia and damaging AI training operations. These strikes were in retaliation for earlier Russian attacks on six of Ukraine's own data centers.
Yandex is commonly described as Russia's Google equivalent and provides a range of digital services to the country, including its primary online search engine, taxi and food delivery apps, e-commerce services, streaming platforms, email, and the GenAI model YandexGPT. The company also operates Yandex Cloud, which hosts digital infrastructure for over 64,000 customers--among these, Yandex claims, are a third of companies on Russia's equivalent of the Fortune 500.
The strikes left Russians unable to make financial transactions, order groceries, and even enter their own homes. The subsequent outages also impaired Russia's gig economy, cutting off taxi drivers and couriers from their earnings. The effects were not merely isolated to Russia, as Kazakhstan, Kyrgyzstan, and Uzbekistan faced similar service interruptions due to their own reliance on Yandex.
The Yandex strikes come in the same year as Iran's drone attacks on Amazon Web Services (AWS) data centers in the Gulf. On March 1, the day after the Iran war began, Iranian drones struck two AWS facilities in the United Arab Emirates and damaged a third in Bahrain. AWS cited "structural damage" and "disrupted power delivery," with the recovery expected to be "prolonged." Ridesharing, financial, and enterprise software services raced to restore their services after they faced outages following the strikes. Months later, AWS was still unable to restore parts of its Gulf data infrastructure. In July, Iran claimed further attacks on Amazon's "data infrastructure" in Bahrain; satellite imagery confirmed damage at two Amazon data centers in Zallaq and Askar.
Q2: Why are data centers becoming targets in war?
A2: Data centers are increasingly being targeted for several reasons:
The infrastructure is essential to multiple sectors and significant GDP growth in a country's economy;
the companies that operate hyperscale data centers often provide critical services themselves to large populations;
governments and militaries are highly reliant on this infrastructure and these services; and
data centers are easy physical targets.
The math is simple. Drones worth tens of thousands of dollars can target billion-dollar data centers, cutting off commercial services critical to the livelihoods of millions of people--and potentially hampering processing and data storage capabilities for militaries. For Iran and Ukraine alike, data centers offered a way to potentially strike a more conventionally powerful adversary's economy and military at once.
Data centers are also physically positioned as ideal targets for airstrikes. Hyperscale data centers are large, immovable buildings with no air defense. Moreover, the proximity of some of these larger data centers to key population and economic hubs, without the potential of significant casualties, makes them strategically impactful without necessarily triggering a disproportionate military response.
Tehran's own justification following the March 2026 strikes was relatively straightforward: The claim was that the facilities were "supporting the enemy's military and intelligence activities." Following this, in April, Iran's Islamic Revolutionary Guard Corps issued a warning that it would target 18 U.S. technology companies in retaliation for subsequent assassinations of Iranian officials; companies named included Microsoft, Google, Apple, Palantir, Nvidia, and Oracle.
Iran's security chief Mohsen Rezaei broadened the state's rationale, naming "cables, data centers, artificial intelligence facilities, and numerous other sites" as targets if fighting resumed. Rezaei claimed around $2 trillion in U.S. digital infrastructure assets existed in the region.
Q3: Who threatens data centers in the United States?
A3: Iran's justification for its strikes in the Gulf shows how an adversary might decide to target data centers in wartime, and the same dual-use logic could apply to facilities in the United States. The Pentagon is increasingly reliant on commercial cloud infrastructure. For example, in 2022 it awarded a $9 billion contract to AWS, Google, Microsoft, and Oracle. Moreover, the National Security Agency has placed some of its data on AWS's cloud under a $10 billion contract, while Microsoft and Oracle operate Pentagon-exclusive cloud regions in Northern Virginia, Iowa, Illinois, and elsewhere. Yet threats to data centers in the United States will likely look very different than those in the Middle East or Europe.
Shielded by geography, U.S. data centers are unlikely to come under attack from missiles or airstrikes. More plausible risks, however, come from individuals in the United States acting on their own grievances or supported by a foreign adversarial government. This threat environment is starting to take shape. Few types of infrastructure draw opposition from across the political spectrum as data centers do, especially as public opposition to the facilities has skyrocketed in recent months and years. A Gallup poll recently found that 71 percent of Americans opposed local data center construction, and voters rejected data centers in 40 out of 41 measures on the ballot in November's midterm elections.
Some of this opposition has manifested in the form of violent threats aimed at data center facilities, personnel, and government officials. While many of the threats investigated by law enforcement have not yielded real-world violence, some individuals have taken concrete steps. For example, police arrested a man in 2021 who actively looked for explosives to destroy an Amazon data center, and another in 2025 who claimed to have acquired materials to burn down an xAI facility. Moreover, the individual implicated in an alleged June 2026 plot to attack the UFC event at the White House had also considered attacks on "AI datacenters, satellite communication facilities, and a California-based technology company," according to the Justice Department.
Domestic extremists may also plot organized attacks. For example, a regional intelligence center in Northern Virginia revealed the potential of antigovernment extremists targeting data centers. A similar warning was issued by a government fusion center in Pennsylvania, naming state-backed actors, criminals, and violent extremists as potential perpetrators. Wired reported that some of these warnings could encompass peaceful protest activities, which has drawn pushback from activists. While the threat exists, it should not be exaggerated, as much of the opposition to data centers remains lawful local activism.
Although these threats currently remain low, foreign adversaries could provide these extremists with organizational, material, or financial resources or use information operations to inflame grievances. Russia has already used this playbook in Europe, recruiting individuals using "gig economy" methods to conduct sabotage activities against critical infrastructure sites through fairly rudimentary means such as arson. RAND Corporation expert Ian Mitch noted that China's growing intelligence network in the United States has moved toward "more violent methods and clandestine tradecraft" that he warns could be directed toward sabotage, though there are few signs of active planning for attacks. Both military and commercial data centers would be natural targets should a crisis emerge, especially as adversaries look to disrupt government and economic activity.
Q4: How vulnerable is a U.S. data center, really?
Many U.S. data centers are highly secured facilities; Microsoft, for example, describes perimeter security, biometrics, security officers, and full-body metal detection, among other interventions, as layered protections at its own data centers. This makes a physical attack inside the facility extremely difficult, though an insider threat in the form of an employee or contractor with legitimate access could face far fewer barriers.
Drones are another matter. While the average American is unlikely to access a Shahed kamikaze drone, commercial drones are inexpensive and widely available. As the Atlantic Council's Bilal Saab noted, many drones are fairly adaptable and difficult to detect, and their primarily civilian use cases make them difficult to regulate. Would-be attackers therefore have convenient tools at their disposal--and have already attempted to use them. For example, in 2020, a modified DJI drone flew toward a Pennsylvania power substation in an apparent attempt to short-circuit it. More recently, in 2024, authorities arrested a white supremacist who allegedly intended to attack an energy facility in Tennessee.
There haven't yet been any reported drone attacks on data centers in the United States. However, limited safeguards could still let an attacker try. As in prior cases targeting energy infrastructure, a commercial drone could be flown toward a data center or the external dependencies that keep it operational, such as substations, external water supply, or fiber-optic cables.
Construction sites present another vulnerability. Data centers under construction are likely to be considerably less secure and more accessible than operational facilities, with fewer layers of protection and a steady flow of third-party contractors, temporary workers, and deliveries. This exposure was demonstrated most recently during a November 6 protest in Amsterdam, when over 100 activists broke into a Microsoft data center construction site. While that breach was a protest rather than an attack, it shows how easily these sites can be accessed, leaving them open to sabotage, arson, theft, or espionage. Moreover, adversaries targeting data center construction sites could also deploy tools to enable spying or conduct drone surveillance of the site.
Q5: What strategies could the United States use to protect itself from physical attacks against data centers?
A5: Currently, data centers are only implicitly a part of the 16 critical infrastructure sectors as defined by the Cybersecurity and Infrastructure Security Agency (CISA). Given their strategic significance, multisectoral dependencies, and unique vulnerabilities, policymakers should move toward an explicit, formal designation as a 17th critical infrastructure sector.
CISA, part of the Department of Homeland Security, defines critical infrastructure as "so vital" that "incapacitation or destruction" would perilously affect U.S. national security, economic security, public health, and/or public safety. While data centers would implicitly fall under the information technology sector (and the communications sector), the lack of clearly defined responsibility limits federal oversight and industry standards to protect the data and processing services that multiple sectors across the country rely on data centers for.
Moreover, local and state law enforcement should more actively coordinate with data center operators on potential physical security risks--engagement that experts argue should happen during data center planning and construction.
As these outlined risks to data centers grow, policymakers need to keep pace with the threat environment, especially as data center growth continues to expand exponentially. These overarching recommendations address the most primary security concerns:
* Data centers should be explicitly classified as a separate sector of critical infrastructure by the Department of Homeland Security.
* Local and state law enforcement should collaborate with data center providers in the early construction stage about realistic risks to physical infrastructure. Data center planning should incorporate information-sharing mechanisms with local and state law enforcement that expand through the continuation of the data center's lifespan.
* States and municipalities where data centers are being built should counter the risk of drone intrusions through local regulation and zoning laws.
* Federally informed, industry-wide standards should be set for hiring employees for the construction and operation of mission-critical data centers, including military and intelligence facilities and hyperscale AI training centers.
As data centers continue to proliferate in the United States, there must be revamped and intentional coordination by both policymakers and industry to invest in building safe, secure facilities that can withstand the potential litany of threats that face this strategic infrastructure.
* * *
A'ndre Gonawela is a fellow with the Strategic Technologies Program at the Center for Strategic and International Studies (CSIS) in Washington, D.C. Taylar Rajic is a fellow with the Strategic Technologies Program at CSIS.
* * *
Original text here: https://www.csis.org/analysis/data-physical-its-security-must-be-too
[Category: ThinkTank]
* * *
Data Is Physical; Its Security Must Be, Too
Opening your apartment door with a fob, paying for your groceries with your phone, ordering an Uber, and scrolling through Instagram don't just exist online; they depend on physical infrastructure. Data centers are the invisible backbone of our highly digitalized daily lives. Apps, websites, and AI tools run on physical ... Show Full Article WASHINGTON, Nov. 18 -- The Center for Strategic and International Studies issued the following Critical Questions Q&A on Nov. 16, 2026, involving A'ndre Gonawela and Taylar Rajic, fellows with the Strategic Technologies Program: * * * Data Is Physical; Its Security Must Be, Too Opening your apartment door with a fob, paying for your groceries with your phone, ordering an Uber, and scrolling through Instagram don't just exist online; they depend on physical infrastructure. Data centers are the invisible backbone of our highly digitalized daily lives. Apps, websites, and AI tools run on physicalservers, and data centers house them, store data, and process the computing tasks that keep those digital services running. If a cluster of data centers were destroyed in Northern Virginia, daily life would essentially be on pause: e-commerce, emails, internet, and essentially any app-based service would not be able to function.
Data centers are increasingly becoming targets in conflict precisely because of their strategic significance. Strikes on data centers have been ongoing in geopolitical conflict in the Middle East and Europe, heavily disrupting essential services during wartime. Although the United States is unlikely to experience the same physical threats as Europe or the Middle East, domestic actors have a different but potentially destructive toolkit at their disposal. The physical threats that are plaguing current geopolitical conflicts--missile strikes and kinetic attacks in active war--are not a concern to U.S.-based data centers. Insider threats and coordinated attacks on facilities, conducted by both domestic actors and foreign-enabled proxies, pose the biggest risks, and are currently severely underappreciated.
Q1: What happened in the attacks on Russia's Yandex data centers? Have we seen these types of strikes before?
A1: From October 8-11, 2026, Ukrainian drones struck three data centers operated by Russian technology company Yandex as part of the ongoing Russia-Ukraine war. Although no deaths were reported, the strikes damaged three of the five data centers operated by the Russian tech giant, seriously disrupting daily life in Russia and damaging AI training operations. These strikes were in retaliation for earlier Russian attacks on six of Ukraine's own data centers.
Yandex is commonly described as Russia's Google equivalent and provides a range of digital services to the country, including its primary online search engine, taxi and food delivery apps, e-commerce services, streaming platforms, email, and the GenAI model YandexGPT. The company also operates Yandex Cloud, which hosts digital infrastructure for over 64,000 customers--among these, Yandex claims, are a third of companies on Russia's equivalent of the Fortune 500.
The strikes left Russians unable to make financial transactions, order groceries, and even enter their own homes. The subsequent outages also impaired Russia's gig economy, cutting off taxi drivers and couriers from their earnings. The effects were not merely isolated to Russia, as Kazakhstan, Kyrgyzstan, and Uzbekistan faced similar service interruptions due to their own reliance on Yandex.
The Yandex strikes come in the same year as Iran's drone attacks on Amazon Web Services (AWS) data centers in the Gulf. On March 1, the day after the Iran war began, Iranian drones struck two AWS facilities in the United Arab Emirates and damaged a third in Bahrain. AWS cited "structural damage" and "disrupted power delivery," with the recovery expected to be "prolonged." Ridesharing, financial, and enterprise software services raced to restore their services after they faced outages following the strikes. Months later, AWS was still unable to restore parts of its Gulf data infrastructure. In July, Iran claimed further attacks on Amazon's "data infrastructure" in Bahrain; satellite imagery confirmed damage at two Amazon data centers in Zallaq and Askar.
Q2: Why are data centers becoming targets in war?
A2: Data centers are increasingly being targeted for several reasons:
The infrastructure is essential to multiple sectors and significant GDP growth in a country's economy;
the companies that operate hyperscale data centers often provide critical services themselves to large populations;
governments and militaries are highly reliant on this infrastructure and these services; and
data centers are easy physical targets.
The math is simple. Drones worth tens of thousands of dollars can target billion-dollar data centers, cutting off commercial services critical to the livelihoods of millions of people--and potentially hampering processing and data storage capabilities for militaries. For Iran and Ukraine alike, data centers offered a way to potentially strike a more conventionally powerful adversary's economy and military at once.
Data centers are also physically positioned as ideal targets for airstrikes. Hyperscale data centers are large, immovable buildings with no air defense. Moreover, the proximity of some of these larger data centers to key population and economic hubs, without the potential of significant casualties, makes them strategically impactful without necessarily triggering a disproportionate military response.
Tehran's own justification following the March 2026 strikes was relatively straightforward: The claim was that the facilities were "supporting the enemy's military and intelligence activities." Following this, in April, Iran's Islamic Revolutionary Guard Corps issued a warning that it would target 18 U.S. technology companies in retaliation for subsequent assassinations of Iranian officials; companies named included Microsoft, Google, Apple, Palantir, Nvidia, and Oracle.
Iran's security chief Mohsen Rezaei broadened the state's rationale, naming "cables, data centers, artificial intelligence facilities, and numerous other sites" as targets if fighting resumed. Rezaei claimed around $2 trillion in U.S. digital infrastructure assets existed in the region.
Q3: Who threatens data centers in the United States?
A3: Iran's justification for its strikes in the Gulf shows how an adversary might decide to target data centers in wartime, and the same dual-use logic could apply to facilities in the United States. The Pentagon is increasingly reliant on commercial cloud infrastructure. For example, in 2022 it awarded a $9 billion contract to AWS, Google, Microsoft, and Oracle. Moreover, the National Security Agency has placed some of its data on AWS's cloud under a $10 billion contract, while Microsoft and Oracle operate Pentagon-exclusive cloud regions in Northern Virginia, Iowa, Illinois, and elsewhere. Yet threats to data centers in the United States will likely look very different than those in the Middle East or Europe.
Shielded by geography, U.S. data centers are unlikely to come under attack from missiles or airstrikes. More plausible risks, however, come from individuals in the United States acting on their own grievances or supported by a foreign adversarial government. This threat environment is starting to take shape. Few types of infrastructure draw opposition from across the political spectrum as data centers do, especially as public opposition to the facilities has skyrocketed in recent months and years. A Gallup poll recently found that 71 percent of Americans opposed local data center construction, and voters rejected data centers in 40 out of 41 measures on the ballot in November's midterm elections.
Some of this opposition has manifested in the form of violent threats aimed at data center facilities, personnel, and government officials. While many of the threats investigated by law enforcement have not yielded real-world violence, some individuals have taken concrete steps. For example, police arrested a man in 2021 who actively looked for explosives to destroy an Amazon data center, and another in 2025 who claimed to have acquired materials to burn down an xAI facility. Moreover, the individual implicated in an alleged June 2026 plot to attack the UFC event at the White House had also considered attacks on "AI datacenters, satellite communication facilities, and a California-based technology company," according to the Justice Department.
Domestic extremists may also plot organized attacks. For example, a regional intelligence center in Northern Virginia revealed the potential of antigovernment extremists targeting data centers. A similar warning was issued by a government fusion center in Pennsylvania, naming state-backed actors, criminals, and violent extremists as potential perpetrators. Wired reported that some of these warnings could encompass peaceful protest activities, which has drawn pushback from activists. While the threat exists, it should not be exaggerated, as much of the opposition to data centers remains lawful local activism.
Although these threats currently remain low, foreign adversaries could provide these extremists with organizational, material, or financial resources or use information operations to inflame grievances. Russia has already used this playbook in Europe, recruiting individuals using "gig economy" methods to conduct sabotage activities against critical infrastructure sites through fairly rudimentary means such as arson. RAND Corporation expert Ian Mitch noted that China's growing intelligence network in the United States has moved toward "more violent methods and clandestine tradecraft" that he warns could be directed toward sabotage, though there are few signs of active planning for attacks. Both military and commercial data centers would be natural targets should a crisis emerge, especially as adversaries look to disrupt government and economic activity.
Q4: How vulnerable is a U.S. data center, really?
Many U.S. data centers are highly secured facilities; Microsoft, for example, describes perimeter security, biometrics, security officers, and full-body metal detection, among other interventions, as layered protections at its own data centers. This makes a physical attack inside the facility extremely difficult, though an insider threat in the form of an employee or contractor with legitimate access could face far fewer barriers.
Drones are another matter. While the average American is unlikely to access a Shahed kamikaze drone, commercial drones are inexpensive and widely available. As the Atlantic Council's Bilal Saab noted, many drones are fairly adaptable and difficult to detect, and their primarily civilian use cases make them difficult to regulate. Would-be attackers therefore have convenient tools at their disposal--and have already attempted to use them. For example, in 2020, a modified DJI drone flew toward a Pennsylvania power substation in an apparent attempt to short-circuit it. More recently, in 2024, authorities arrested a white supremacist who allegedly intended to attack an energy facility in Tennessee.
There haven't yet been any reported drone attacks on data centers in the United States. However, limited safeguards could still let an attacker try. As in prior cases targeting energy infrastructure, a commercial drone could be flown toward a data center or the external dependencies that keep it operational, such as substations, external water supply, or fiber-optic cables.
Construction sites present another vulnerability. Data centers under construction are likely to be considerably less secure and more accessible than operational facilities, with fewer layers of protection and a steady flow of third-party contractors, temporary workers, and deliveries. This exposure was demonstrated most recently during a November 6 protest in Amsterdam, when over 100 activists broke into a Microsoft data center construction site. While that breach was a protest rather than an attack, it shows how easily these sites can be accessed, leaving them open to sabotage, arson, theft, or espionage. Moreover, adversaries targeting data center construction sites could also deploy tools to enable spying or conduct drone surveillance of the site.
Q5: What strategies could the United States use to protect itself from physical attacks against data centers?
A5: Currently, data centers are only implicitly a part of the 16 critical infrastructure sectors as defined by the Cybersecurity and Infrastructure Security Agency (CISA). Given their strategic significance, multisectoral dependencies, and unique vulnerabilities, policymakers should move toward an explicit, formal designation as a 17th critical infrastructure sector.
CISA, part of the Department of Homeland Security, defines critical infrastructure as "so vital" that "incapacitation or destruction" would perilously affect U.S. national security, economic security, public health, and/or public safety. While data centers would implicitly fall under the information technology sector (and the communications sector), the lack of clearly defined responsibility limits federal oversight and industry standards to protect the data and processing services that multiple sectors across the country rely on data centers for.
Moreover, local and state law enforcement should more actively coordinate with data center operators on potential physical security risks--engagement that experts argue should happen during data center planning and construction.
As these outlined risks to data centers grow, policymakers need to keep pace with the threat environment, especially as data center growth continues to expand exponentially. These overarching recommendations address the most primary security concerns:
* Data centers should be explicitly classified as a separate sector of critical infrastructure by the Department of Homeland Security.
* Local and state law enforcement should collaborate with data center providers in the early construction stage about realistic risks to physical infrastructure. Data center planning should incorporate information-sharing mechanisms with local and state law enforcement that expand through the continuation of the data center's lifespan.
* States and municipalities where data centers are being built should counter the risk of drone intrusions through local regulation and zoning laws.
* Federally informed, industry-wide standards should be set for hiring employees for the construction and operation of mission-critical data centers, including military and intelligence facilities and hyperscale AI training centers.
As data centers continue to proliferate in the United States, there must be revamped and intentional coordination by both policymakers and industry to invest in building safe, secure facilities that can withstand the potential litany of threats that face this strategic infrastructure.
* * *
A'ndre Gonawela is a fellow with the Strategic Technologies Program at the Center for Strategic and International Studies (CSIS) in Washington, D.C. Taylar Rajic is a fellow with the Strategic Technologies Program at CSIS.
* * *
Original text here: https://www.csis.org/analysis/data-physical-its-security-must-be-too
[Category: ThinkTank]
CSIS Issues Commentary: What a New Biosecurity Policy Can Teach Us About Overseeing Frontier AI
WASHINGTON, Nov. 18 -- The Center for Strategic and International Studies issued the following commentary:
* * *
What a New Biosecurity Policy Can Teach Us About Overseeing Frontier AI
Commentary by Melissa Hopkins
November 16, 2026
This month, the White House released a long-awaited policy to strengthen the bioeconomy and biosecurity of the United States. The policy--which covers nucleic acid synthesis screening (NASS, sometimes also referred to as gene synthesis screening)--arrives as states and Congress grapple with various approaches to frontier AI governance and are increasingly focused ... Show Full Article WASHINGTON, Nov. 18 -- The Center for Strategic and International Studies issued the following commentary: * * * What a New Biosecurity Policy Can Teach Us About Overseeing Frontier AI Commentary by Melissa Hopkins November 16, 2026 This month, the White House released a long-awaited policy to strengthen the bioeconomy and biosecurity of the United States. The policy--which covers nucleic acid synthesis screening (NASS, sometimes also referred to as gene synthesis screening)--arrives as states and Congress grapple with various approaches to frontier AI governance and are increasingly focusedon leveraging independent verification organizations (IVOs) to oversee frontier AI companies. The history of NASS policy in the United States--and the new policy's dual-verification approach--offers hard-won lessons for where to go next on AI policy.
The White House Replaces Self-Attestation with Verified Compliance
Scientists across the United States often don't make their own nucleic acids in-house. Instead, they purchase synthetic DNA and RNA from companies that provide this as a service, known commonly as providers. For over a decade, a consortium of companies known as the International Gene Synthesis Consortium (IGSC) voluntarily screened synthetic nucleic acid orders coming in from scientists to ensure that their orders were not unnecessarily dual-use or that the person making the order had a legitimate reason for purchasing a particular sequence that could be dual-use. The group formed in response to high-profile biosecurity episodes such as the 2006 purchase of a smallpox DNA fragment from a provider without adequate vetting and the reconstruction of the 1918 pandemic influenza virus from synthetic DNA. The IGSC saw the potential of their technology to be misused and opted to ensure that this would not happen, both for the safety and security of humanity and for survival of their business model. If a horrific biosecurity incident occurred using DNA or RNA that a provider sold, not only could the company face potentially catastrophic liability, but it also would lose consumer trust and invite reactive and overly burdensome regulatory requirements.
But over time, this voluntary regime proved not to be enough. While companies in the IGSC represented a large share of the nucleic acid provider industry, bad actors could simply order from a company that did not screen. Red teaming of IGSC member companies by private actors, whereby individuals made orders that should not have been completed or at least flagged under IGSC policies, also revealed that the member companies did not necessarily comply with their own voluntary policies.
The result has been an evolution of NASS policy from self-regulation into escalating levels of verified compliance. In the process of that shift, NASS policy has shown itself not only as one of the few emerging technology topics that represent a rare convergence of bipartisan political will, industry support, and technical readiness, but also as an exemplar for how to approach other advanced technologies, including AI. Advocates for the new NASS policy point to strong nucleic acid synthesis security as the first step in a multilayered defense approach to biosecurity to deter misuse without hindering beneficial research, due to its unique effectiveness as a choke point.
The White House's newly released policy, the National Policy for Nucleic Acid Synthesis Screening, replaces the Biden-era 2024 Framework for Nucleic Acid Synthesis Screening. On full implementation, the new policy will also supersede the Biden-era 2023 HHS Screening Framework Guidance (which itself began during President Trump's first term). The policy is a triumph for advocates who have sought to build on the 2024 framework. Where the 2024 framework required all federally funded entities to purchase their synthetic nucleic acids only from providers that screened orders, that policy relied on providers self-attesting to the screening policies outlined in the 2024 framework. The updated policy goes further than the 2024 framework in numerous ways, but most importantly, it establishes a trusted vendor program through the National Institute of Standards and Technology (NIST) and extends its reach to subrecipients of federal funding.
Instead of providers self-attesting that they comply with specified screening practices, the new policy tasks NIST with issuing screening standards that third-party screeners will then need to demonstrate compliance with to become verified by NIST. Providers that demonstrate use of such screeners or can demonstrate their own compliance with the standards will then be approved as "adherent suppliers," and a public list of such companies will be maintained for all federally funded recipients and subrecipients to look to when purchasing synthetic nucleic acids. This effectively ends the voluntary NASS regime by incentivizing companies that want access to the U.S. nucleic acid synthesis market--which is about 75 percent federally funded and is the largest in the world--to actually comply with screening standards rather than simply claim to.
The new policy covers more providers by expanding the reach of the policy to subrecipients of federal funding. For example, if an academic lab receives federal funding, the lab and any of its subcontractors would also be bound by the policy. However, it's unclear how exactly this will be implemented: In mixed funding situations, for example, where an academic lab gets both private and federal funding, is it still allowed to purchase synthetic nucleic acids from noncompliant providers using private funds, or does the federal funding requirement extend to the use of any funds, regardless of whether federal funds are being used explicitly or not? This same question would similarly extend to subrecipients of federal funds.
Frontier AI Needs the Same Two Layers of Verification
Frontier AI governance is different in several ways from biosecurity; for example, it is moving much faster than developments in the NASS policy. While NASS policy advocates have been working for about two decades, advocates for frontier AI governance emerged more or less around 2016, the same time as the breakthrough in transformer models that underpin the most advanced modern AI systems. Frontier AI is also far less reliant on federal funding. Still, it has striking similarities to the unfolding of NASS policy.
Whereas the IGSC was formed three years after concerning biosecurity incidents, the Frontier Model Forum--which works on developing similar industry best practices--was formed in 2023, prior to any such comparable incidents. (There have now been several AI-enabled cybersecurity incidents--e.g., the UK AI Security Institute incident and the OpenAI-Hugging Face incident--that are arguably comparable to, or worse than, those biosecurity incidents.) In 2024, synthetic nucleic acid providers still self-attested to best practices; several weeks ago, key frontier AI companies signed the White House Accord on Super Intelligence that served as their own form of self-attestation. In that document, frontier AI companies agreed to implement stronger AI safety and security practices such as strong monitoring and control of AI as well as using third parties to carry out "independent assessments of whether the controls, monitoring, and detection are operating as intended." Frontier AI governance has effectively mirrored the evolution of NASS policy on an accelerated timeline, rapidly moving from an approach that simply issued standards and included industry-led trade associations to a new mix of self-attestation and third-party verification.
In practice, that accelerated pace offers a mixed prognosis. In many ways, the pace at which frontier AI governance is developing, compared to the pace of NASS governance, is encouraging. At the same time, the jump in capabilities for frontier AI is much faster than the pace of NASS progress. So, when compared more closely, it could be that the ratio between governance and capabilities is roughly the same across both fields, explaining the slower governance in NASS and the faster governance in frontier AI. In some ways that does not matter. Given the scope and scale of AI's impacts, it is essential to borrow as much as possible from existing policy solutions that deal with similar kinds of problems, so that policymakers don't have to waste time reinventing the wheel or bringing stakeholders on board.
The new NASS policy demonstrates that not only was self-attestation of screening not enough for the safety and security of the United States (and, via the nature of biology, the world), but that self-attestation of the use of third-party screeners was also not enough. Without a verification regime like the one NIST is spinning up for NASS, policymakers do not have answers to critical questions about the competencies of the third parties that the frontier labs are using--nor do the frontier labs necessarily even have to use these third parties in the first place.
Indeed, the key strength of the verification regime established by the new policy is that there are two layers of verification: first for the screeners (or third parties), and a second for the adherent suppliers (who must use verified screeners). The policy even sets out an ambitious timeline for this: 90 days from publication for standards, and 160 days for the first list of adherent suppliers. Though frontier AI capabilities are reportedly accelerating, similar timelines seem reasonable to develop standards for AI safety and security that balance speed to market with enough robustness to ensure they do not immediately become irrelevant upon issuance.
Establishing a formal NASS-like regime for AI would also ensure not only that frontier AI companies are using third parties for AI safety and security, but that those third parties adhere to a specific standard that the government has developed with expert and stakeholder input. Without that, voluntary use of third parties for safety and security risks serving as safety theater. For example, a frontier AI company could use a third-party auditor or evaluator whose practices may allow for additional incidents to occur or simply forum shop for the third party with the laxest practices to tick off the compliance box.
With a new Congress slated to begin its work in a few months--filled with lawmakers more attuned to AI policy issues--dual-layer verification for frontier AI should be considered a key pillar in any frontier AI governance approach. Policymakers don't need to make mistakes in one domain that they have already made in another, and they need to act with speed given rapidly advancing frontier AI capabilities.
Congress Should Adopt the NASS Model for Frontier AI
A federal funding lever for frontier AI governance likely won't have the same effectiveness as it does for NASS policy because providers' and frontier AI companies' customer bases are so different. But there are already several other comparable levers that policymakers can use to incentivize frontier AI companies to participate in a dual-verification regime.
There are several bills on the state and federal levels that take a self-regulatory approach to third-party verification. California's SB 813, enacted in September 2026, tasks a California agency with designating IVOs, third-party verifiers for generative AI. The bill encourages the agency to look to existing regulatory and audit standards rather than create new ones that the IVOs would need to adhere to, contrary to the NASS policy. The bill also explicitly states that it does not require AI companies to use IVOs. Illinois' SB 0315, enacted in July, goes further: IVOs are mandatory for frontier AI companies. Here, too, the third party itself has no requirements and is not held to any standards beyond minimizing conflicts of interest.
It's true that auditing a frontier AI company for its adherence to its own safety and security practices is different from doing the work of implementing safety and security practices themselves, as is the case with NASS screeners. It's part of why AI IVO policies, which are also being considered on the federal level as part of the FRONTIER Act, fail in enhancing safety and security the same way that the new NASS policy does on each of the layers of verification. On the first layer of verification for a screener, the NASS policy asks, "Does this screener implement practices collectively agreed upon by the government and expert stakeholders that decrease biosecurity risk?"; on the second layer it asks, "Does this provider use a screener that has been verified to implement such practices, or implement such practices themselves?" Instead, the AI policies under consideration are asking something more akin to, "Does this Provider use any third party to test their own practices?" That is certainly better than nothing, but insufficient given the lessons from NASS policy.
Given ongoing state-level appetite for AI legislation, one key policy lever in the new year could be dual verification for frontier AI companies implemented in new state laws. The shortcoming of this approach, however, is that states lack the resources needed to publish high-quality standards. Without the standards, there can be no underlying accountability framework for frontier AI companies and their third-party verifiers.
A more practical approach would be for Congress to pass a bill that mirrors the Trump administration's approach to the NASS policy. Such legislation should (1) require NIST to publish standards within 90 days on AI safety and security that mirror the commitments made in the White House Accord on Super Intelligence; (2) task NIST with verifying which third parties can adhere to such standards; and (3) require frontier AI companies to use verified third parties for AI safety and security within 160 days (or be verified by NIST to adhere to them, themselves).
Absent federal legislation, another option would be executive action on federal procurement--specifically, mandating that no AI companies can do business with federal agencies or federally funded entities unless they comply with the dual-verification regime established by the White House (similar to the new NASS policy). The benefit of this approach is that it would not rely on an act of Congress, but the drawback is that the market for frontier AI is much different from that for synthetic nucleic acids and is far less reliant on federal contracts, so it is not clear how much such a policy would actually incentivize compliance.
All these approaches to frontier AI governance have trade-offs, but the lesson from NASS policy is clear: Third-party verification only enhances safety and security if the third parties themselves are held to a standard. NASS policy took nearly two decades of voluntary commitments and self-attestation to arrive at this conclusion, but policymakers don't have the same luxury of time for frontier AI. In the new year, Congress and other policymakers should start where NASS policy has ended, rather than repeating the mistakes made getting there.
* * *
Melissa Hopkins is an adjunct fellow (non-resident) at the Center for Strategic and International Studies in Washington, D.C., and an emerging technology advisor at the Johns Hopkins Center for Health Security.
* * *
Original text here: https://www.csis.org/analysis/what-new-biosecurity-policy-can-teach-us-about-overseeing-frontier-ai
[Category: ThinkTank]
* * *
What a New Biosecurity Policy Can Teach Us About Overseeing Frontier AI
Commentary by Melissa Hopkins
November 16, 2026
This month, the White House released a long-awaited policy to strengthen the bioeconomy and biosecurity of the United States. The policy--which covers nucleic acid synthesis screening (NASS, sometimes also referred to as gene synthesis screening)--arrives as states and Congress grapple with various approaches to frontier AI governance and are increasingly focused ... Show Full Article WASHINGTON, Nov. 18 -- The Center for Strategic and International Studies issued the following commentary: * * * What a New Biosecurity Policy Can Teach Us About Overseeing Frontier AI Commentary by Melissa Hopkins November 16, 2026 This month, the White House released a long-awaited policy to strengthen the bioeconomy and biosecurity of the United States. The policy--which covers nucleic acid synthesis screening (NASS, sometimes also referred to as gene synthesis screening)--arrives as states and Congress grapple with various approaches to frontier AI governance and are increasingly focusedon leveraging independent verification organizations (IVOs) to oversee frontier AI companies. The history of NASS policy in the United States--and the new policy's dual-verification approach--offers hard-won lessons for where to go next on AI policy.
The White House Replaces Self-Attestation with Verified Compliance
Scientists across the United States often don't make their own nucleic acids in-house. Instead, they purchase synthetic DNA and RNA from companies that provide this as a service, known commonly as providers. For over a decade, a consortium of companies known as the International Gene Synthesis Consortium (IGSC) voluntarily screened synthetic nucleic acid orders coming in from scientists to ensure that their orders were not unnecessarily dual-use or that the person making the order had a legitimate reason for purchasing a particular sequence that could be dual-use. The group formed in response to high-profile biosecurity episodes such as the 2006 purchase of a smallpox DNA fragment from a provider without adequate vetting and the reconstruction of the 1918 pandemic influenza virus from synthetic DNA. The IGSC saw the potential of their technology to be misused and opted to ensure that this would not happen, both for the safety and security of humanity and for survival of their business model. If a horrific biosecurity incident occurred using DNA or RNA that a provider sold, not only could the company face potentially catastrophic liability, but it also would lose consumer trust and invite reactive and overly burdensome regulatory requirements.
But over time, this voluntary regime proved not to be enough. While companies in the IGSC represented a large share of the nucleic acid provider industry, bad actors could simply order from a company that did not screen. Red teaming of IGSC member companies by private actors, whereby individuals made orders that should not have been completed or at least flagged under IGSC policies, also revealed that the member companies did not necessarily comply with their own voluntary policies.
The result has been an evolution of NASS policy from self-regulation into escalating levels of verified compliance. In the process of that shift, NASS policy has shown itself not only as one of the few emerging technology topics that represent a rare convergence of bipartisan political will, industry support, and technical readiness, but also as an exemplar for how to approach other advanced technologies, including AI. Advocates for the new NASS policy point to strong nucleic acid synthesis security as the first step in a multilayered defense approach to biosecurity to deter misuse without hindering beneficial research, due to its unique effectiveness as a choke point.
The White House's newly released policy, the National Policy for Nucleic Acid Synthesis Screening, replaces the Biden-era 2024 Framework for Nucleic Acid Synthesis Screening. On full implementation, the new policy will also supersede the Biden-era 2023 HHS Screening Framework Guidance (which itself began during President Trump's first term). The policy is a triumph for advocates who have sought to build on the 2024 framework. Where the 2024 framework required all federally funded entities to purchase their synthetic nucleic acids only from providers that screened orders, that policy relied on providers self-attesting to the screening policies outlined in the 2024 framework. The updated policy goes further than the 2024 framework in numerous ways, but most importantly, it establishes a trusted vendor program through the National Institute of Standards and Technology (NIST) and extends its reach to subrecipients of federal funding.
Instead of providers self-attesting that they comply with specified screening practices, the new policy tasks NIST with issuing screening standards that third-party screeners will then need to demonstrate compliance with to become verified by NIST. Providers that demonstrate use of such screeners or can demonstrate their own compliance with the standards will then be approved as "adherent suppliers," and a public list of such companies will be maintained for all federally funded recipients and subrecipients to look to when purchasing synthetic nucleic acids. This effectively ends the voluntary NASS regime by incentivizing companies that want access to the U.S. nucleic acid synthesis market--which is about 75 percent federally funded and is the largest in the world--to actually comply with screening standards rather than simply claim to.
The new policy covers more providers by expanding the reach of the policy to subrecipients of federal funding. For example, if an academic lab receives federal funding, the lab and any of its subcontractors would also be bound by the policy. However, it's unclear how exactly this will be implemented: In mixed funding situations, for example, where an academic lab gets both private and federal funding, is it still allowed to purchase synthetic nucleic acids from noncompliant providers using private funds, or does the federal funding requirement extend to the use of any funds, regardless of whether federal funds are being used explicitly or not? This same question would similarly extend to subrecipients of federal funds.
Frontier AI Needs the Same Two Layers of Verification
Frontier AI governance is different in several ways from biosecurity; for example, it is moving much faster than developments in the NASS policy. While NASS policy advocates have been working for about two decades, advocates for frontier AI governance emerged more or less around 2016, the same time as the breakthrough in transformer models that underpin the most advanced modern AI systems. Frontier AI is also far less reliant on federal funding. Still, it has striking similarities to the unfolding of NASS policy.
Whereas the IGSC was formed three years after concerning biosecurity incidents, the Frontier Model Forum--which works on developing similar industry best practices--was formed in 2023, prior to any such comparable incidents. (There have now been several AI-enabled cybersecurity incidents--e.g., the UK AI Security Institute incident and the OpenAI-Hugging Face incident--that are arguably comparable to, or worse than, those biosecurity incidents.) In 2024, synthetic nucleic acid providers still self-attested to best practices; several weeks ago, key frontier AI companies signed the White House Accord on Super Intelligence that served as their own form of self-attestation. In that document, frontier AI companies agreed to implement stronger AI safety and security practices such as strong monitoring and control of AI as well as using third parties to carry out "independent assessments of whether the controls, monitoring, and detection are operating as intended." Frontier AI governance has effectively mirrored the evolution of NASS policy on an accelerated timeline, rapidly moving from an approach that simply issued standards and included industry-led trade associations to a new mix of self-attestation and third-party verification.
In practice, that accelerated pace offers a mixed prognosis. In many ways, the pace at which frontier AI governance is developing, compared to the pace of NASS governance, is encouraging. At the same time, the jump in capabilities for frontier AI is much faster than the pace of NASS progress. So, when compared more closely, it could be that the ratio between governance and capabilities is roughly the same across both fields, explaining the slower governance in NASS and the faster governance in frontier AI. In some ways that does not matter. Given the scope and scale of AI's impacts, it is essential to borrow as much as possible from existing policy solutions that deal with similar kinds of problems, so that policymakers don't have to waste time reinventing the wheel or bringing stakeholders on board.
The new NASS policy demonstrates that not only was self-attestation of screening not enough for the safety and security of the United States (and, via the nature of biology, the world), but that self-attestation of the use of third-party screeners was also not enough. Without a verification regime like the one NIST is spinning up for NASS, policymakers do not have answers to critical questions about the competencies of the third parties that the frontier labs are using--nor do the frontier labs necessarily even have to use these third parties in the first place.
Indeed, the key strength of the verification regime established by the new policy is that there are two layers of verification: first for the screeners (or third parties), and a second for the adherent suppliers (who must use verified screeners). The policy even sets out an ambitious timeline for this: 90 days from publication for standards, and 160 days for the first list of adherent suppliers. Though frontier AI capabilities are reportedly accelerating, similar timelines seem reasonable to develop standards for AI safety and security that balance speed to market with enough robustness to ensure they do not immediately become irrelevant upon issuance.
Establishing a formal NASS-like regime for AI would also ensure not only that frontier AI companies are using third parties for AI safety and security, but that those third parties adhere to a specific standard that the government has developed with expert and stakeholder input. Without that, voluntary use of third parties for safety and security risks serving as safety theater. For example, a frontier AI company could use a third-party auditor or evaluator whose practices may allow for additional incidents to occur or simply forum shop for the third party with the laxest practices to tick off the compliance box.
With a new Congress slated to begin its work in a few months--filled with lawmakers more attuned to AI policy issues--dual-layer verification for frontier AI should be considered a key pillar in any frontier AI governance approach. Policymakers don't need to make mistakes in one domain that they have already made in another, and they need to act with speed given rapidly advancing frontier AI capabilities.
Congress Should Adopt the NASS Model for Frontier AI
A federal funding lever for frontier AI governance likely won't have the same effectiveness as it does for NASS policy because providers' and frontier AI companies' customer bases are so different. But there are already several other comparable levers that policymakers can use to incentivize frontier AI companies to participate in a dual-verification regime.
There are several bills on the state and federal levels that take a self-regulatory approach to third-party verification. California's SB 813, enacted in September 2026, tasks a California agency with designating IVOs, third-party verifiers for generative AI. The bill encourages the agency to look to existing regulatory and audit standards rather than create new ones that the IVOs would need to adhere to, contrary to the NASS policy. The bill also explicitly states that it does not require AI companies to use IVOs. Illinois' SB 0315, enacted in July, goes further: IVOs are mandatory for frontier AI companies. Here, too, the third party itself has no requirements and is not held to any standards beyond minimizing conflicts of interest.
It's true that auditing a frontier AI company for its adherence to its own safety and security practices is different from doing the work of implementing safety and security practices themselves, as is the case with NASS screeners. It's part of why AI IVO policies, which are also being considered on the federal level as part of the FRONTIER Act, fail in enhancing safety and security the same way that the new NASS policy does on each of the layers of verification. On the first layer of verification for a screener, the NASS policy asks, "Does this screener implement practices collectively agreed upon by the government and expert stakeholders that decrease biosecurity risk?"; on the second layer it asks, "Does this provider use a screener that has been verified to implement such practices, or implement such practices themselves?" Instead, the AI policies under consideration are asking something more akin to, "Does this Provider use any third party to test their own practices?" That is certainly better than nothing, but insufficient given the lessons from NASS policy.
Given ongoing state-level appetite for AI legislation, one key policy lever in the new year could be dual verification for frontier AI companies implemented in new state laws. The shortcoming of this approach, however, is that states lack the resources needed to publish high-quality standards. Without the standards, there can be no underlying accountability framework for frontier AI companies and their third-party verifiers.
A more practical approach would be for Congress to pass a bill that mirrors the Trump administration's approach to the NASS policy. Such legislation should (1) require NIST to publish standards within 90 days on AI safety and security that mirror the commitments made in the White House Accord on Super Intelligence; (2) task NIST with verifying which third parties can adhere to such standards; and (3) require frontier AI companies to use verified third parties for AI safety and security within 160 days (or be verified by NIST to adhere to them, themselves).
Absent federal legislation, another option would be executive action on federal procurement--specifically, mandating that no AI companies can do business with federal agencies or federally funded entities unless they comply with the dual-verification regime established by the White House (similar to the new NASS policy). The benefit of this approach is that it would not rely on an act of Congress, but the drawback is that the market for frontier AI is much different from that for synthetic nucleic acids and is far less reliant on federal contracts, so it is not clear how much such a policy would actually incentivize compliance.
All these approaches to frontier AI governance have trade-offs, but the lesson from NASS policy is clear: Third-party verification only enhances safety and security if the third parties themselves are held to a standard. NASS policy took nearly two decades of voluntary commitments and self-attestation to arrive at this conclusion, but policymakers don't have the same luxury of time for frontier AI. In the new year, Congress and other policymakers should start where NASS policy has ended, rather than repeating the mistakes made getting there.
* * *
Melissa Hopkins is an adjunct fellow (non-resident) at the Center for Strategic and International Studies in Washington, D.C., and an emerging technology advisor at the Johns Hopkins Center for Health Security.
* * *
Original text here: https://www.csis.org/analysis/what-new-biosecurity-policy-can-teach-us-about-overseeing-frontier-ai
[Category: ThinkTank]
CSIS Issues Commentary Entitled 'Dealing With Canada'
WASHINGTON, Nov. 18 -- The Center for Strategic and International Studies issued the following commentary:
* * *
Dealing with Canada
Commentary by William Alan Reinsch
November 17, 2026
Trade with Canada is once again in the news thanks to Democratic promises in the wake of the elections to repeal Trump's tariffs on the country. The latest polling data shows that all the tariffs are wildly unpopular, probably because they have become part of the affordability debate, and the ones on Canada are least popular of all. While repeal sentiment is strongest in the border states, it also shows up ... Show Full Article WASHINGTON, Nov. 18 -- The Center for Strategic and International Studies issued the following commentary: * * * Dealing with Canada Commentary by William Alan Reinsch November 17, 2026 Trade with Canada is once again in the news thanks to Democratic promises in the wake of the elections to repeal Trump's tariffs on the country. The latest polling data shows that all the tariffs are wildly unpopular, probably because they have become part of the affordability debate, and the ones on Canada are least popular of all. While repeal sentiment is strongest in the border states, it also shows upin states where Canadian parts and components are part of local supply chains, most notably the auto industry, and areas that have been the beneficiaries of Canadian tourism and "snowbirds" who come south for the winter. Once commonly sighted in Florida and Arizona, they have become rare birds indeed.
So, the stage seems to be set for a repeal effort once the new Congress convenes in January and Democrats are in the majority. However, there are several obstacles to success. First, action may prove unnecessary depending on what happens between now and the end of the year. Increased 50 percent tariffs on many Canadian goods are slated to go into effect January 1, which creates an incentive for Canada to make a deal, which was probably the tariffs' purpose. The election results may also help persuade the U.S. administration that a deal makes political sense. Agreement has been elusive so far, though Canadian prime minister Mark Carney sounded optimistic earlier this month when he noted that the United States was "asking less and offering more," but he also said the parties were not there yet. If a deal is struck, however, that will take some of the steam out of the repeal debate, and at the very least will make the construction of a congressional coalition harder.
Second, while it has been easy for Democrats to say the tariffs should be removed, they have been vague about precisely which tariffs. Unfortunately, the tariffs Canada would most like to see disappear are the Section 232 tariffs on steel and aluminum, which are the very tariffs a good number of Democrats would like to keep. Remember that Senator-elect Sherrod Brown (D-OH) is returning to the Senate. He has long been a strong supporter of those tariffs. I see no indication he has changed his mind, and I expect others to follow his lead. That may not be a majority, but it will be enough to make Senate passage more complicated, particularly if Republicans decide to unite in resistance; 60 votes will be needed to overcome a filibuster, and Democrats will have at most 56. The Republicans have a larger decision to make in the new year: whether they will simply resist all Democratic initiatives or try to find areas where the two parties can cooperate and produce something. The latter is preferable by far, and a decision on how to respond to tariff repeal might be an early test.
Third, tariff repeal means a revenue reduction, which, in turn, means a search for "pay-fors"--revenue-raising measures that will offset the loss. It is difficult to separate tariff revenue by country in real time, but it appears that in 2025 Canada paid approximately $10 billion in tariffs. The 2026 amount is hard to assess but may be more than that. This is not an insurmountable hurdle, but it will complicate the path of any legislation.
Finally, and most important, Democrats will have to construct a veto-proof majority, because Trump is certain to veto a tariff repeal bill. That means 290 votes in the House and 67 in the Senate, assuming everyone votes. Democrats are short of that in both bodies and will need to rely on Republican defections. My experience on the Hill taught me that getting members of the president's party to go against him on a veto override vote is extraordinarily difficult. Members are under intense pressure from their own leadership and from the White House to support the president. That can often lead to deals in which the White House doles out goodies to wavering members. A slightly easier path would be to tuck tariff repeal into an unrelated "must-pass" bill such as the defense authorization or an appropriation. That would also attract a veto, but it might attract a larger coalition willing to override.
Thus, the path to victory is complicated despite the popularity of the idea. This is also an opportunity to remind people of the dangers of short-term thinking. Right now, there is a lot of commentary on the dire straits of the Republican Party, but the people writing it off now are the same people who wrote off the Democrats two years ago. The pendulum swings, and it always swings back eventually. A vote on tariff relief for Canada will be an early test of the ability of Democrats to mobilize their troops and demonstrate unity, and it will be an opportunity for Republicans to decide whether they want to break with Trump on some issues or fall on their swords supporting him.
* * *
William A. Reinsch is a senior adviser (non-resident) and Scholl Chair emeritus with the Economics Program and Scholl Chair in International Business at the Center for Strategic and International Studies in Washington, D.C. He can be reached at wreinsch7@gmail.com.
* * *
Original text here: https://www.csis.org/analysis/dealing-canada
[Category: ThinkTank]
* * *
Dealing with Canada
Commentary by William Alan Reinsch
November 17, 2026
Trade with Canada is once again in the news thanks to Democratic promises in the wake of the elections to repeal Trump's tariffs on the country. The latest polling data shows that all the tariffs are wildly unpopular, probably because they have become part of the affordability debate, and the ones on Canada are least popular of all. While repeal sentiment is strongest in the border states, it also shows up ... Show Full Article WASHINGTON, Nov. 18 -- The Center for Strategic and International Studies issued the following commentary: * * * Dealing with Canada Commentary by William Alan Reinsch November 17, 2026 Trade with Canada is once again in the news thanks to Democratic promises in the wake of the elections to repeal Trump's tariffs on the country. The latest polling data shows that all the tariffs are wildly unpopular, probably because they have become part of the affordability debate, and the ones on Canada are least popular of all. While repeal sentiment is strongest in the border states, it also shows upin states where Canadian parts and components are part of local supply chains, most notably the auto industry, and areas that have been the beneficiaries of Canadian tourism and "snowbirds" who come south for the winter. Once commonly sighted in Florida and Arizona, they have become rare birds indeed.
So, the stage seems to be set for a repeal effort once the new Congress convenes in January and Democrats are in the majority. However, there are several obstacles to success. First, action may prove unnecessary depending on what happens between now and the end of the year. Increased 50 percent tariffs on many Canadian goods are slated to go into effect January 1, which creates an incentive for Canada to make a deal, which was probably the tariffs' purpose. The election results may also help persuade the U.S. administration that a deal makes political sense. Agreement has been elusive so far, though Canadian prime minister Mark Carney sounded optimistic earlier this month when he noted that the United States was "asking less and offering more," but he also said the parties were not there yet. If a deal is struck, however, that will take some of the steam out of the repeal debate, and at the very least will make the construction of a congressional coalition harder.
Second, while it has been easy for Democrats to say the tariffs should be removed, they have been vague about precisely which tariffs. Unfortunately, the tariffs Canada would most like to see disappear are the Section 232 tariffs on steel and aluminum, which are the very tariffs a good number of Democrats would like to keep. Remember that Senator-elect Sherrod Brown (D-OH) is returning to the Senate. He has long been a strong supporter of those tariffs. I see no indication he has changed his mind, and I expect others to follow his lead. That may not be a majority, but it will be enough to make Senate passage more complicated, particularly if Republicans decide to unite in resistance; 60 votes will be needed to overcome a filibuster, and Democrats will have at most 56. The Republicans have a larger decision to make in the new year: whether they will simply resist all Democratic initiatives or try to find areas where the two parties can cooperate and produce something. The latter is preferable by far, and a decision on how to respond to tariff repeal might be an early test.
Third, tariff repeal means a revenue reduction, which, in turn, means a search for "pay-fors"--revenue-raising measures that will offset the loss. It is difficult to separate tariff revenue by country in real time, but it appears that in 2025 Canada paid approximately $10 billion in tariffs. The 2026 amount is hard to assess but may be more than that. This is not an insurmountable hurdle, but it will complicate the path of any legislation.
Finally, and most important, Democrats will have to construct a veto-proof majority, because Trump is certain to veto a tariff repeal bill. That means 290 votes in the House and 67 in the Senate, assuming everyone votes. Democrats are short of that in both bodies and will need to rely on Republican defections. My experience on the Hill taught me that getting members of the president's party to go against him on a veto override vote is extraordinarily difficult. Members are under intense pressure from their own leadership and from the White House to support the president. That can often lead to deals in which the White House doles out goodies to wavering members. A slightly easier path would be to tuck tariff repeal into an unrelated "must-pass" bill such as the defense authorization or an appropriation. That would also attract a veto, but it might attract a larger coalition willing to override.
Thus, the path to victory is complicated despite the popularity of the idea. This is also an opportunity to remind people of the dangers of short-term thinking. Right now, there is a lot of commentary on the dire straits of the Republican Party, but the people writing it off now are the same people who wrote off the Democrats two years ago. The pendulum swings, and it always swings back eventually. A vote on tariff relief for Canada will be an early test of the ability of Democrats to mobilize their troops and demonstrate unity, and it will be an opportunity for Republicans to decide whether they want to break with Trump on some issues or fall on their swords supporting him.
* * *
William A. Reinsch is a senior adviser (non-resident) and Scholl Chair emeritus with the Economics Program and Scholl Chair in International Business at the Center for Strategic and International Studies in Washington, D.C. He can be reached at wreinsch7@gmail.com.
* * *
Original text here: https://www.csis.org/analysis/dealing-canada
[Category: ThinkTank]
