Anthropic has circulated a draft federal AI safety bill to US lawmakers that would give the government limited authority to block or suspend the deployment of advanced artificial intelligence models that pose an imminent risk of catastrophic harm.
The proposal is not yet law. According to authorities, Anthropic representatives privately presented the draft to lawmakers on Capitol Hill. The company confirmed that the bill reflects recommendations in its Advanced AI Framework, which it has been discussing with federal and state policymakers.
Table of Contents
What the Draft Bill Would Do
Under the reported proposal, the federal government could seek to block an AI model’s release or use only when the model presents a foreseeable and imminent risk of more than $1 billion in damage or the death or serious injury of more than 50 people.
The risk would have to be connected to specific circumstances, including the creation or release of a biological weapon, a cyberattack, or an AI system evading controls established by its developer.
The proposal would focus on advanced or “frontier” AI models rather than every artificial intelligence product. Anthropic’s official framework describes frontier models as systems capable of creating serious biological, cybersecurity, loss-of-control, or automated research and development risks.
Court Approval Would Be Required
The draft would place the proposed enforcement process under judicial oversight. Reuters reported that the Commerce Department could ask a federal judge to halt a model after determining that it presents an applicable imminent risk.
A judge in the US District Court for the District of Columbia would have authority to order a suspension under the reported framework. The proposal would also prevent other federal departments, agencies, or officers from independently suspending a model’s development, deployment, or internal use.
Anthropic said the proposal would give the federal government the ability to “block or deter dangerous deployments of the most powerful models,” while limiting the circumstances in which that power could be used.
Testing and Independent Evaluations
The proposed legislation would establish new obligations for developers of the most powerful AI systems. These obligations would include testing models for serious safety risks, publishing summaries of the findings, undergoing independent evaluations, and maintaining a security program to protect models and related infrastructure.
According to Reuters, independent evaluators could be registered with the Commerce Department. If an evaluator identified an “imminent enumerated risk,” the evaluator would reportedly have to notify the department within 72 hours after completing the assessment.
The draft would allow AI laboratories to pay independent evaluators, provided that payment was not tied to the outcome of the evaluation. That provision is likely to raise questions about how evaluator independence would be protected in practice.
Possible Effect on State AI Laws
The draft also addresses the relationship between federal AI regulation and state laws. The reported proposal would preempt some state AI safety requirements relating to frontier-model risks.
Federal preemption could create a single national standard for AI developers operating across the United States. It could also restrict states from adopting different rules in areas covered by the federal framework.
Anthropic’s official Advanced AI Framework takes a narrower position on preemption. The company says Congress should not broadly override state law unless the federal protections are at least as strong as the proposed framework. It also says states should retain authority over areas such as child safety and consumer protection that are outside the specific federal safety rules.
Which Companies Would Be Covered?
Anthropic’s framework is aimed at large companies developing the most advanced AI systems. It proposes applying the rules to models trained using more than 10²⁵ floating-point operations, or FLOPs, when they are developed by companies earning more than $500 million in AI-related revenue or spending more than $1 billion on AI research and development.
Those thresholds would determine which developers fall within the proposed federal AI safety requirements. The final bill could change the thresholds or use a different method to identify frontier AI developers.
Proposal Comes as Congress Considers AI Regulation
The draft comes as the United States continues to debate whether and how to adopt a comprehensive federal AI law.
The debate also reflects broader questions about how artificial intelligence will affect legal work, courts, and professional responsibility. For more context, read our analysis of how AI may change the future of legal careers.
Current discussions have focused on AI safety, cybersecurity, transparency, model evaluations, consumer protection, and the division of authority between federal and state governments.
Anthropic’s framework argues that transparency alone may not be enough to address the risks posed by the most capable AI models. It calls for government enforcement authority, independent testing, public reporting, and stronger security requirements.
The proposal has also drawn comparisons with other AI safety legislation. Charlie Bullock of the Institute for Law & AI described it as a reasonable proposal but said it was weaker than the competing FRONTIER Act, according to The Hindu’s report on the draft.
What Happens Next?
The proposal remains a draft and has not been enacted or formally established as federal law. Its language could change if lawmakers decide to introduce legislation, and the bill would still face committee review, debate, amendments, and votes in Congress.
The proposal is likely to contribute to the wider debate over federal AI regulation, judicial review, government authority, and the future role of state AI laws. Whether Congress adopts Anthropic’s approach or develops a different framework remains uncertain.
This article is for general information only and is not legal advice. The proposal described above is subject to change and has not been enacted as federal law.
