WASHINGTON, October 16, 2026 — Artificial intelligence company Anthropic has presented US lawmakers with a draft bill aimed at establishing federal safeguards for advanced AI systems, proposing a legal framework that would allow government intervention when a model presents a narrowly defined risk of catastrophic harm.
The proposal, reported by Politico, would establish new requirements for testing, independent evaluation and disclosure while limiting the circumstances under which authorities could suspend the release or use of an AI model. The initiative comes as Washington faces growing pressure to determine how powerful AI technologies should be regulated without unnecessarily restricting innovation.
Under the proposed legislation, government intervention would be permitted when a model presents a foreseeable and imminent risk of causing more than $1 billion in damage or killing or seriously injuring more than 50 people. The specified risks would need to arise from scenarios involving biological weapons, an AI system escaping its developer’s controls or a cyberattack.
The draft has not been enacted into law, and its provisions remain proposals rather than binding federal requirements.
Anthropic Proposed Rules Set a High Threshold for Government Intervention
The legislation would establish a defined process for addressing the most serious potential threats associated with frontier AI models, which are among the most advanced systems developed by leading technology companies.
Under the proposed framework, the US Department of Commerce would play a central role in overseeing safety assessments and responding to findings that a model could pose an imminent, specified danger.
The department could ask a federal judge in the District of Columbia to order a suspension of a model if the relevant legal threshold were met. The proposal would restrict other federal departments, agencies and government officials from independently suspending AI development, deployment or internal use under the framework.
This approach would place judicial oversight at the centre of decisions involving the temporary restriction of advanced AI systems. It would also limit the circumstances in which the government could intervene, rather than granting regulators broad authority to halt a model based on general safety concerns.
The proposed damage and casualty thresholds would establish measurable criteria for intervention. However, determining whether a particular AI system presents an imminent risk would require technical assessments and evidence about its capabilities, potential uses and the likelihood of harmful outcomes.
The framework therefore seeks to establish a formal relationship between technical safety evaluations and government enforcement.
Independent Assessments Would Become a Central Requirement
A Anthropic significant element of the proposal is the introduction of independent evaluators to assess AI developers and their most advanced models.
Under the draft, evaluators would register with the Commerce Department and periodically examine companies’ systems. These assessments would be intended to identify potential risks that internal testing might not adequately reveal.
AI developers would be permitted to pay for these evaluations, provided that compensation was not linked to the outcome. The arrangement would seek to allow external technical scrutiny while reducing the incentive for evaluators to produce favourable conclusions.
The proposal would also require evaluators who identify a qualifying imminent risk to report their findings to the Commerce Department within 72 hours of completing the assessment.
Such a requirement could create a formal reporting channel between independent technical specialists and federal authorities. It would also place additional responsibilities on AI companies to cooperate with external assessments and maintain appropriate safety documentation.
However, questions remain about how evaluators would be selected, how conflicts of interest would be managed and what technical standards would determine whether a model crosses the proposed risk threshold.
The effectiveness of the system would depend partly on the independence of assessments, the quality of available evidence and the government’s ability to respond promptly when credible warnings emerge.
Federal Regulation Could Reshape State-Level Rules
The proposal arrives amid continuing disagreements over the appropriate division of responsibility between federal and state authorities in regulating artificial intelligence.
The draft would override certain state AI laws, potentially creating a more uniform national framework for companies operating across the United States. Supporters of a federal approach have argued that consistent rules could reduce regulatory fragmentation and make compliance easier for developers working across multiple jurisdictions.
However, federal pre-emption could also limit states’ ability to introduce additional protections in response to local concerns. The extent of that effect would depend on the final legislation and the specific state requirements it superseded.
The Anthropic debate reflects a broader policy challenge: establishing consistent standards for a rapidly developing technology while preserving meaningful safeguards against foreseeable harm.
Anthropic has previously advocated for stronger safeguards for advanced AI systems. A company spokesperson said the proposed bill reflected recommendations outlined in the company’s Advanced AI Framework, which sets out its approach to managing risks associated with increasingly capable models.
The US Department of Commerce had not immediately responded to a request for comment on the draft.
Safety Debate Intensifies as AI Capabilities Expand
The proposal emerges as concerns about AI security move beyond theoretical risks to include questions about autonomous software, cybersecurity and the ability of developers to maintain control over increasingly capable systems.
Advanced AI tools can perform complex tasks with limited human intervention, creating potential benefits in research, software development and business operations. The same capabilities have also prompted scrutiny of whether existing safeguards are sufficient to prevent misuse or unintended consequences.
For policymakers, the challenge is to distinguish ordinary operational risks from scenarios that could justify extraordinary government intervention.
The proposed legislation would create a legal mechanism for addressing a limited category of severe threats, while leaving broader questions about transparency, accountability and the commercial deployment of AI open to continued debate.
Its future will depend on whether lawmakers support the proposed thresholds, independent assessment requirements and limits on executive intervention. Until legislation is formally introduced, debated and enacted, the proposal remains one possible approach to federal AI governance rather than an established regulatory framework.
Also Read :- Anthropic IPO Filing Reveals Deep Dependence on Big Tech Partners
