The White House's Gold Eagle Initiative Scales Back, the AI Kill Switch Act, and Bill Gates' Warnings and Proposals

Table of Contents

1. Gold Eagle under Pressure: Narrowed Implementation and Expanding Regulatory Debate

2. "AI Kill Switch Act" Proposed in Response to Hacks by Rogue Models

3. Bill Gates Warns of Risks Posed by AI and Proposes Measures to Address Them

4. Court Strikes Down Pentagon's Anthropic Blacklisting

5. FSB Chair Warns of Frontier AI Risks to Financial Stability

6. US Secures G20 Agreement on Light-Touch AI Regulation

7. Trump Administration Backs OpenAI in Landmark Copyright Case, Citing National Security

8. Bipartisan "Doctors Not AI Act" Would Require Human Review of Health Insurance Denials

9. NAIC Releases AI Risk Evaluation Supplement 5.0

10. Google's WeatherNext Cyclones Model Advances AI Hurricane Forecasting

This month's briefing also covers the court ruling against the Pentagon's blacklisting of Anthropic, as well as the Financial Stability Board chair's warnings about frontier AI risks. Meanwhile, the bipartisan Doctors Not AI Act would prohibit AI from making denials of health insurance coverage based on medical necessity. Read on for a deeper dive into these and more key updates.

Regulatory, Legislative & Litigation Developments

Gold Eagle under Pressure: Narrowed Implementation and Expanding Regulatory Debate

Since our last update, the Trump administration's Gold Eagle cybersecurity clearinghouse has been substantially scaled back. A Cybersecurity and Infrastructure Security Agency (CISA) fact sheet issued August 14, 2026, recast the program as a triage and deduplication layer feeding the existing Vulnerability Information and Coordination Environment (VINCE) platform, rather than the broader coordination system described in July. The Executive Order 14409 benchmark for "covered frontier models," due August 1, appeared to slip, though the White House later stated the framework was completed on schedule without disclosing its contents or its reviewers.

Commentators have questioned the Treasury Department's coordination authority, CERT/CC's (Computer Emergency Response Team Coordination Center) capacity to absorb AI-scale reporting volume, and the risk of concentrating disclosures in a single platform that could itself become a target. Meanwhile, Senator Warner's Secure AI Development Act (S.5061) would replace the EO's voluntary access model with a mandatory, penalty-backed National Security Agency (NSA) review.

Most immediately, the CISA 2015 liability protections that underpin voluntary vulnerability-sharing expire September 30 absent congressional action. Stakeholders participating in, or considering, Gold Eagle's voluntary reporting mechanisms should track both the sunset and the pending bill, either of which could materially shift the risk calculus for disclosure.

"AI Kill Switch Act" Proposed in Response to Hacks by Rogue Models

In response to hacks by rogue artificial intelligence agents, Representatives Ted Lieu (D-California) and Nathaniel Moran (R-Texas) introduced a bill, the AI Kill Switch Act (the Kill Switch Act), that would require companies that operate artificial intelligence models to have the capability to shut down their models and to suspend or terminate access for accounts, users, or activity patterns that pose risks to the integrity of the models.

The bill further provides that if an entity uses one of these capabilities, it must file a report with the secretary of the Department of Homeland Security within 15 days and permits the secretary to impose fines of up to $20 million per day against entities who fail to comply with an emergency order issued by the secretary, and up to $2 million per day for otherwise violating the Kill Switch Act. If signed into law, the Kill Switch Act has the potential to incentivize companies to police the use of their models and ensure the safety of their users, while empowering the secretary to order corrective action upon determining that a covered incident has occurred.

Bill Gates Warns of Risks Posed by AI and Proposes Measures to Address Them

On August 26, 2026, Bill Gates published an essay on his website warning of three major risks posed by AI: (1) massive job loss as humans are no longer needed for many tasks; (2) the development of weapons, including bombs, bioweapons, and computer viruses, as well as AI-enabled fraud, disinformation, and cyberattacks on critical infrastructure; and (3) negative impacts on child and adolescent development, including the deterioration of relational and critical thinking skills.

However, Gates does not believe that the development of AI should be halted, because he recognizes that AI has the potential to provide significant benefits in health care, agriculture, education, government services, and the development of small businesses. Accordingly, he proposes three ways to mitigate AI's risks: (1) a "token tax" on AI and robots that finances job re-training; (2) the development of domestic and international frameworks to regulate AI and set priorities across agencies and address cross-border threats; and (3) the designation of certain jobs, including those that require relational skills, as "Human Reserved."

Although Gates recognizes his framework will take a significant amount of time and resources to develop and implement — and it may not be adopted — at a minimum, his essay has already served to spark conversations about the need to develop a coherent, thoughtful approach to AI, which will undoubtedly become even more pressing as its adoption accelerates.

Court Strikes Down Pentagon's Anthropic Blacklisting

The Northern District of California ruled on August 27, 2026, that the Department of Defense acted unlawfully when Defense Secretary Pete Hegseth designated AI company Anthropic a "supply chain risk," a label which previously was only used for entities tied to foreign adversaries. The court found that the designation amounted to unconstitutional retaliation against Anthropic for exercising its First Amendment rights, specifically its insistence on maintaining safety guardrails that prevented its Claude AI models from being used in fully autonomous lethal weapons or mass surveillance of Americans.

In the Court's opinion, Judge Rita Lin characterized the government's actions as "illegal and baseless," concluding that the Pentagon misinterpreted the governing statute and failed to follow required procedures in imposing the designation. Judge Lin further stated that "[t]he empty invocation of national security is not a blank check to punish and retaliate against government critics." The ruling effectively bars the Pentagon from enforcing the supply chain risk label, which had threatened to exclude Anthropic products from the entire defense contracting ecosystem and jeopardized hundreds of millions of dollars in existing contracts. The ruling was stayed for seven days to allow the Department of Justice to file an appeal, and a separate Anthropic lawsuit challenging a related designation remains pending in Washington, D.C.

FSB Chair Warns of Frontier AI Risks to Financial Stability

In an August 28, 2026, letter to G20 Finance Ministers and Central Bank Governors, Financial Stability Board Chair Andrew Bailey warned that frontier AI models' increasingly sophisticated autonomy, problem-solving abilities, and threat capabilities are complicating the financial-stability risk landscape.

Bailey identified cyber risk as the most immediate concern, noting that frontier AI could materially alter the speed, scale, and economics of cyber threats and potentially undermine confidence across the financial system, particularly given reliance on concentrated third-party technology providers. Because the global financial system is highly interconnected, AI-enabled cyber disruptions could spread across jurisdictions through common providers, shared infrastructure, and cross-border financial activity.

Bailey also observed that many jurisdictions lack protocols governing the development, release, and deployment of advanced frontier models and he called for globally coordinated steps to support their safe and responsible deployment.

US Secures G20 Agreement on Light-Touch AI Regulation

At the G20 Innovation Ministerial meeting in Chapel Hill, North Carolina, on September 1 and 2, 2026, the United States pressed fellow G20 members to adopt a hands-off approach to AI regulation. White House technology adviser Michael Kratsios introduced the "Carolina Principles," which call on governments to reserve new regulation for genuinely novel AI issues that cannot be addressed by existing laws, rather than treating every emerging technology as a first-of-its-kind policy problem.

G20 members unanimously endorsed the principles, though the agreement is a nonbinding framework commitment rather than a binding international instrument. The US position contrasts sharply with the European Union's approach, which has moved forward with enforceable requirements for high-risk AI systems under its AI Act. The framework is expected to shape the broader AI governance debate heading into the G20 Leaders' Summit in Miami in December 2026.

Trump Administration Backs OpenAI in Landmark Copyright Case, Citing National Security

On September 1, 2026, the US Department of Justice filed a 20-page statement of interest in the consolidated AI copyright litigation pending in the US District Court for the Southern District of New York supporting OpenAI in its copyright disputes with the New York Times and other publishers, marking the first time the federal government has formally weighed in on copyright litigation involving AI training. The statement argued that training large language models on copyrighted texts is "highly transformative" and qualifies as fair use, drawing a distinction between the training process itself and any potentially infringing outputs. The DOJ warned that "[r]ules of law that make it significantly more difficult to develop a robust AI industry in the United States therefore threaten national security and give a competitive advantage to foreign adversaries who are not so encumbered." Associate Attorney General Stanley Woodward stated that "AI dominance is critical to promote national security, prosperity, and economic mobility for all Americans."

Bipartisan "Doctors Not AI Act" Would Require Human Review of Health Insurance Denials

On September 1, 2026, Representatives Greg Landsman (D-Ohio), Buddy Carter (R-Georgia), Kim Schrier, M.D. (D-Washington), and Tom Barrett (R-Michigan) introduced the bipartisan Doctors Not AI Act (H.R. 10210), which would prohibit AI from making or issuing health insurance coverage denials based on medical necessity. The bill would allow insurers to continue using AI to assist with claims processing, but would require that any denial based on medical necessity be made by a licensed health care professional who independently reviews the patient's individual medical circumstances. The legislation would also require insurers to disclose when AI is used in the claims review process and includes provisions to prevent AI from unfairly denying or limiting mental health and substance use disorder care.

NAIC Releases AI Risk Evaluation Supplement 5.0

On August 31, 2026, the National Association of Insurance Commissioners' Big Data and AI Working Group released for public comment its AI Risk Evaluation Supplement (formerly the AI Systems Evaluation Tool) version 5.0, which is intended to be used in connection with market conduct and financial exams. The supplement is being developed in connection with a 12-state pilot program running from March through September 2026. Among other changes, version 5.0 expands the scope of the supplement to include generalized linear models (GLMs), noting "[a]s with other machine learning techniques, GLMs are not without risk of causing unfair discrimination or other adverse consumer outcomes and require effective governance over data quality, model development, validation, implementation, ongoing performance monitoring, and compliance with applicable laws and regulations." The comment period ends on September 29, 2026. Version 5.0 is an intermediate step in a multi-version process; the NAIC plans to release a further revised version for a second comment period before considering adoption of the supplement at the Fall National Meeting in November 2026.

Google's WeatherNext Cyclones Model Advances AI Hurricane Forecasting

Historically, meteorologists and organizations that monitored weather conditions across the globe struggled to develop accurate hurricane forecasts and models. Over time, traditional weather forecasting tools became relatively adept at predicting a developed storm's path, but those tools often failed to provide advanced modeling regarding storm intensity. The same was true with respect to AI forecasting models. Early AI models became proficient at predicting a storm's track, but were much less reliable at predicting its intensity. Google recently announced advancements in its new WeatherNext Cyclones model that move AI hurricane forecasting a giant step forward.

The WeatherNext Cyclones model combines global atmospheric data with observations from nearly 5,000 historical tropical cyclones to improve both track and intensity forecasts. Google says the model provides more than a full day of additional predictive accuracy for a cyclone's track, intensity, and wind structure — meaning its three-day forecasts are roughly as accurate as previous models' two-day forecasts, which represents an improvement equivalent to about a decade of conventional meteorological progress. Google also said that the model can generate up to 1,000 possible storm scenarios, giving forecasters a probability distribution rather than a single predicted path. During Hurricane Melissa in 2025, Google says the model helped identify rapid intensification and a likely Jamaica landfall five days in advance, potentially giving forecasters substantially more time to warn the public and prepare for the storm.