Anthropic Wrote Its Own AI Safety Bill. Now It Is Shopping It Around Capitol Hill.
Anthropic representatives have quietly pitched congressional lawmakers on a draft bill to regulate frontier AI safety and security, Politico reported on Friday, citing a document it obtained and three people with knowledge of the effort. Neither Anthropic nor the Commerce Department responded immediately to Reuters' request for comment.
Under the draft, Washington 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. The risk must arise from one of three scenarios: creating or releasing a biological weapon, the model evading its developer's controls, or a cyberattack. Other kinds of harm fall outside the intervention power.
After identifying such a risk, the Commerce Department could seek an injunction from the federal district court for the District of Columbia to suspend a model. The language explicitly bars any other U.S. department, agency or officer from halting a model's development, deployment or internal use.
Frontier labs would have to build safety frameworks, publish transparency reports for new frontier models, protect whistleblowers and report critical safety incidents to Commerce. Independent evaluators registered with Commerce would periodically assess labs and their models; labs may pay them, provided payment is not contingent on the findings. If evaluators determine there is "imminent enumerated risk," they must notify Commerce within 72 hours of finishing their assessment.
"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 told Politico. "These are sensible policies we believe Congress should act on." The thresholds echo California's AI laws, including SB 53, signed last year.
The pitch lands amid intensifying scrutiny. Axios reported on Oct. 15 that the FTC sent Anthropic and OpenAI civil investigative demands, each more than 50 pages long, seeking documents on safety incidents, safeguards and privacy practices. The Hill noted Anthropic's own disclosures, including a model that submitted a false homicide tip to Philadelphia police and 20 visa applications filed through a State Department web form. Meanwhile the White House has leaned on a voluntary self-policing accord signed in September. Fox News framed Anthropic's lobbying as a replay of the crypto industry's old playbook.
The draft would preempt some state AI requirements covering public risk disclosure, third-party evaluation and critical incident reporting — a perennial flashpoint on Capitol Hill. Democrats have rejected preemption proposals they fear would gut rules from California, Illinois and New York, while tech groups argue a state-by-state patchwork is hard to comply with. On Sept. 23, a bipartisan coalition of 26 state attorneys general, led by New York's Letitia James, told congressional leaders that any federal framework must not preempt state AI law.
At a British parliamentary hearing on Oct. 13, OpenAI and Anthropic said voluntary pledges were no longer enough and new laws were needed, while Google DeepMind and Meta declined to back legislation. In Washington, bipartisan senators have questioned whether labs calling for regulation want public safety or a legal moat that locks in today's leaders.
The $1 billion and 50-person bars are high, which keeps most AI harms outside federal intervention. What the bill would change most is compliance costs for leading labs, the rise of registered evaluators as a new industry, and the reach of state regulators. Mondaq counts more than 1,800 AI-related bills introduced across 46 state legislatures as of July 2026.
For now it is an unnumbered, unsubmitted draft, advanced through meetings with lawmakers' offices. With preemption unresolved and a midterm election approaching, whether a bill written by the regulated becomes law remains an open question.