Key Notes
- Anthropic’s draft bill would allow intervention only if a frontier model poses a foreseeable, imminent risk of more than $1 billion in damage or more than 50 deaths or serious injuries.
- Reported trigger pathways are biological weapons, a model evading its developer’s controls, or a cyberattack.
- The proposed process would put registered independent evaluators and a federal court between a lab and any forced halt, and it is not yet law.
Anthropic has been circulating a draft federal AI-safety bill on Capitol Hill that would give the US government a narrow new route to block or deter the deployment of powerful AI models, according to reporting by Politico and Reuters. The proposal sets a deliberately high intervention threshold: a model would have to pose a foreseeable, imminent risk of more than $1 billion in damage, or of killing or seriously injuring more than 50 people.
That is not a proposal to regulate every chatbot or software feature. It is a framework aimed at the most capable frontier systems, and it would tie any government action to a small set of catastrophic-risk pathways. The draft matters because it sketches how one of the leading AI labs believes Washington should gain enforcement power while limiting the scope of that power.
What would trigger a government intervention
Reuters reported that the threshold would apply only where the risk stems from the creation or release of a biological weapon, a model evading its developer’s controls, or a cyberattack. The bill would not require a catastrophe to happen first. Rather, independent evaluators and the government would have to establish that the risk is foreseeable and imminent before a court could intervene.
The numerical thresholds are significant because they set a high legal bar. A harmful output, an isolated security failure or ordinary consumer harm would not by itself meet the standard described in the reporting. The draft appears designed to focus federal intervention on rare events with mass-casualty or system-wide economic consequences, while keeping lower-level safety, consumer-protection and state-law questions outside that emergency mechanism.
Anthropic has already outlined that policy direction publicly. Its Advanced AI Framework calls for government authority to block or deter dangerous deployments, alongside testing, transparency, independent evaluation and security requirements for frontier developers. A spokesperson told Reuters that the draft bill reflects those earlier proposals.
Evaluators would sit between labs and the government
Under the reported design, independent evaluators would register with the Department of Commerce and periodically assess both frontier labs and their models. Labs could pay those evaluators, provided payment is not linked to the outcome of an assessment. If an evaluator found an “imminent enumerated risk,” it would have to report that finding to Commerce within 72 hours of completing its work.
Commerce would not simply switch off a model on its own. The department could ask a federal judge in the District of Columbia to halt a model’s development, deployment or use after making the required risk finding, Reuters reported. The language would also prohibit other US departments, agencies or officers from independently suspending development, deployment or internal use under that mechanism.
That structure is meant to make the process more formal than a voluntary company policy while avoiding a general-purpose executive power over AI. It also raises practical questions that lawmakers would need to answer if the draft becomes legislation: what evidence proves an imminent model risk, how much access evaluators need, who qualifies as independent, and how a court weighs uncertainty against the cost of stopping a system.
A high threshold in an unsettled policy debate
The draft sits alongside a growing set of proposals that try to define “catastrophic” AI risk before an incident occurs. The existing FRONTIER Act also uses a more-than-50-casualty or $1 billion threshold in its definition of catastrophic risk. Anthropic’s reported proposal adds a specific judicial path for blocking a model and safeguards intended to constrain which federal officials can seek that outcome.
Supporters of this approach can argue that it gives regulators a tool for the kind of biosecurity, cyber and loss-of-control risks that are hard to address after deployment. Critics are likely to question whether the thresholds are so high that meaningful intervention comes too late, or whether a framework written with input from leading closed-model labs could entrench their position over smaller developers and open-weight projects.
The evaluator system is another point of tension. We recently covered Anthropic’s expanded access for independent safety researchers, a voluntary arrangement with a different purpose and scope. Turning independent review into part of a federal enforcement process would make evaluator independence, technical standards and due process central policy questions rather than internal governance choices.
It is a draft, not a new law
Nothing in the reporting indicates that the proposal has been introduced as a bill or enacted. It is a draft being discussed with lawmakers, and Congress could rewrite its thresholds, powers and coverage or decide not to move it forward. The Commerce Department did not immediately respond to Reuters’ request for comment.
Even so, the document clarifies the kind of compromise Anthropic is pursuing: limited federal power to stop only the most dangerous deployments, combined with mandatory testing, transparency and independent review for the biggest frontier developers. Whether that is a workable guardrail or an overly narrow safety valve will be decided in a public policy fight that is only beginning.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
