Anthropic Bill Gives State Power to Block AI Releases

Anthropic presented US lawmakers with a draft bill that would let the government block an AI model’s release if officials determined it posed an imminent risk of killing or injuring more than 50 people, or causing more than $1 billion in damages. The proposal puts state power over a model’s release at the center of the debate.

Who Gets to Decide

The draft sets a high threshold: the threat must involve more than 50 people killed or injured, or more than $1 billion in damage. But the proposal also places the decision in government hands. Officials would determine whether a risk is imminent and could then stop the model’s release.

That’s the mechanism Anthropic put before lawmakers. The proposal doesn’t describe a public decision-making process; instead, it gives the government the stated authority. The draft hasn’t become law. It remains a proposal presented to lawmakers, not an enacted rule.

The bill names three ways a model could pose the specified threat: creating or releasing a biological weapon, going beyond human control, or enabling a cyberattack. Those conditions define the cases in which the government could block a release. They also leave unanswered how authorities would assess an imminent threat.

Politico first reported the proposal. The intervention is framed as AI safety and security, but its practical lever is a government veto over release. The immediate issue isn’t only what a model might do; it’s who gets the power to decide whether the model may reach the public.

Threat Reports, Privacy and Unequal Scrutiny

Separately, legal experts questioned whether existing US law requires AI companies to report threats of violence. Duties to warn can apply to mental health professionals or parties with a special relationship, they said, but those duties may not cover AI companies. That uncertainty sits alongside the draft bill’s more direct proposal for government control over model releases.

The question has been raised by the February 10, 2026, mass shooting in Tumbler Ridge, British Columbia, and by lawsuits alleging OpenAI failed to prevent it. The lawsuits make allegations, while legal experts debate what existing law requires.

Experts also raised concerns about privacy, false threat reports and unequal monitoring. They pointed to how hard it can be to distinguish a credible threat from fictional writing, research or intrusive thoughts. Treating those different things as interchangeable could turn ambiguous language into a report. Unequal monitoring raises another question: who faces scrutiny, and who decides what counts as danger?

The article describes no grassroots response, mutual-aid effort or direct action around the proposal. Instead, it records an institutional response: a company presenting legislation to lawmakers, and legal experts examining duties and risks under existing law. At the top of the process, Anthropic proposes government power to block releases; at the level of individuals, experts warn that privacy, false reports and unequal monitoring belong to the same debate. Neither the draft’s specific thresholds nor the legal uncertainty settles those concerns.