Anthropic Pitches AI Safety Bill to Lawmakers

Anthropic put a draft bill before US lawmakers this week, asking for government power to block an AI model's release if it poses an imminent danger. Politico first reported the proposal. It would let officials stop a model if they decide it could kill or injure more than 50 people, or cause more than $1 billion in damages. It's a striking request from a company that builds the technology itself, and it arrives as worries about the pace of AI development grow louder.

What the Bill Would Cover

The threat would need to come from one of three sources: a model helping create or release a biological weapon, an AI system slipping beyond human control, or a cyberattack. The categories are narrow by design. The draft targets catastrophe, not the everyday harms regulators already struggle to police. Still, the thresholds matter. A model that could injure fifty people or destroy a billion dollars in value isn't a hypothetical concern for the families and communities left to absorb the fallout.

The Gap in Existing Law

A second question is gaining force, and it has nothing to do with bioweapons. Legal experts have questioned whether existing US law requires AI companies to report threats of violence at all. The February 10, 2026, mass shooting in Tumbler Ridge, British Columbia, and lawsuits alleging that OpenAI failed to prevent it have brought that question to the fore. Families bringing those suits ask a basic question: when a company's product helps form part of the chain leading to a killing, who must spot the warning signs?

The answer, according to legal experts quoted in the reporting, may be no one. Current duties to warn can apply to mental health professionals or parties with a special relationship to a potential victim. Those duties may not reach AI companies, which occupy an unusual middle ground: they hold enormous amounts of private conversation, but don't fall within professional frameworks that say who must speak up when someone is in danger.

The Tradeoffs Are Real

Experts said there’s no easy fix. They raised concerns about privacy, false threat reports, and unequal monitoring that could fall hardest on already over-policed communities. They also noted how hard it can be to distinguish a credible threat from fictional writing, academic research, or intrusive thoughts that many people have but never act on. A system that flags too much harms ordinary users. One that flags too little leaves people exposed.

The debate makes clear that companies largely make these decisions in private, while democratic institutions haven't set the rules. Whether Congress takes up Anthropic's proposal or writes something different will determine whether the public gets a say in how officials strike those tradeoffs.

Why This Matters: The proposal and lawsuits point to the same underlying gap. Powerful AI systems now touch people's daily lives, yet the legal duties meant to protect the public from danger were written for doctors, therapists, and other professionals—not software that talks with millions of people every day. The Tumbler Ridge shooting, which killed people in a British Columbia community in February 2026, has made that gap impossible to ignore. Families who lost loved ones are asking courts to decide what a company owed them, and legal experts say current US law offers no clear answer. Anthropic's bill would give government a narrow emergency power to halt releases that threaten mass harm, giving the public oversight of a technology that has largely governed itself. Whether that oversight is adequate, and whether it protects ordinary people as well as it guards against catastrophe, rests with lawmakers balancing safety, privacy, and fairness. The stakes are very high for communities that bear the consequences of these decisions, and they deserve a seat at the table.