Anthropic Quietly Pitched Its Own AI Safety Bill to Congress: Government Could Halt a Model Only if It Risks 50 Deaths or $1 Billion in Damages

Anthropic has been quietly shopping its own AI safety legislation to Capitol Hill. According to a Politico report published October 16, representatives of the San Francisco AI lab pitched lawmakers on a confidential draft bill to regulate frontier AI safety and security — an effort confirmed by three people with knowledge of the private deliberations. Anthropic did not deny it: "This bill represents the proposals outlined in Anthropic's Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June," a spokesperson said, calling the provisions "sensible policies we believe Congress should act on."

The draft sets a deliberately high bar for government intervention. Regulators could block a model's release or use only if it poses a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously injuring more than 50 people — and the risk must stem from one of three enumerated scenarios: the creation or release of a biological weapon, the model evading its developer's controls, or a cyberattack. The language mirrors California's SB 53, signed into law last year, which Anthropic publicly supported.

Enforcement would run through a registry of independent evaluators. Third-party assessment organizations would register with the Commerce Department and periodically evaluate frontier labs and their models; labs would be allowed to pay their own evaluators, so long as payment is not contingent on the results. If evaluators find an "imminent enumerated risk," they must report it to Commerce within 72 hours of completing the assessment. Commerce could then ask a judge — specifically, a federal judge for the District of Columbia — to order a model suspended, and the bill explicitly prohibits any other US department, agency or officer from halting a model's development, deployment or internal use.

The draft would also preempt some state AI laws covering public disclosure of risk, third-party evaluation and critical-incident reporting — a political flashpoint, since Democrats have repeatedly rejected preemption proposals they feared would neuter standards passed by California, Illinois and New York. The bill additionally requires frontier labs to maintain safety frameworks, publish transparency reports for new frontier models, protect whistleblowers and report critical safety incidents to Commerce.

Context matters for why Anthropic wants a statutory process. Over the summer, the company was subject to sweeping export controls that shut down two of its most advanced models — a decision made, according to Politico, in a whirlwind series of phone calls between administration officials and Anthropic executives, which unsettled much of the industry. A defined legal threshold and a single judicial choke point would replace discretionary cutoffs with something predictable. Notably, none of the agent incidents labs disclosed this summer — models probing websites, platforms and even production systems — came close to the bill's thresholds of mass casualties or ten-figure damage.

The proposal lands in a crowded field. In the House, Representatives Lori Trahan and Jay Obernolte have advanced the FRONTIER Act, which would let the government stop deployments of risky models. In the Senate, Majority Leader John Thune with Amy Klobuchar and Ted Cruz have been negotiating a catastrophic-risk bill, over objections from Senator Maria Cantwell, who wants a larger government review role; further left, Representative Greg Casar and Senator Bernie Sanders have called for a pause on frontier development and a ban on superintelligence.

What the draft really signals is a lab trying to shape the rules that will govern it — before someone else does. A company-founded threshold that requires billion-dollar or mass-casualty harm before intervention is, critics will note, a very company-friendly line. Supporters will counter that an enforceable, narrow, court-checked process is more realistic than the alternatives now circulating. Either way, one of the industry's leading safety advocates has moved from publishing frameworks to drafting statute.