Anthropic Pitches Its Own AI Safety Bill to Congress, With a $1 Billion or 50-Death Bar for Blocking AI Models

Anthropic has quietly pitched Capitol Hill lawmakers on draft legislation to regulate the safety of frontier AI models, Politico reported on October 16, citing three people familiar with the effort. The draft would let the federal government block the release or use of an AI model only when it poses a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously injuring more than 50 people.

An Anthropic spokesperson told Reuters that the bill reflects the proposals in the company’s Advanced AI Framework, which it has been advocating with federal and state lawmakers since June. The spokesperson said the bill would give the government the “ability to block or deter dangerous deployments of the most powerful models,” while requiring developers to test frontier AI models, be transparent about their findings, and submit to independent evaluation. An Anthropic spokesperson separately told Politico: “These are sensible policies we believe Congress should act on.”

Background: a lab writing its own rules

Politico described the pitch as a “notable effort” by an AI lab to guide the rules covering its own technology. It lands in a crowded field of AI safety proposals. According to PYMNTS, Anthropic’s approach is consistent with the Artificial Intelligence Risk Management and Security Act of 2026, introduced by a trio of Senate Democrats. A separate bill from Senator Maria Cantwell would direct the National Institute of Standards and Technology to develop standards against catastrophic AI harms, while Senator Josh Hawley’s AI Risk Evaluation Act would put the Department of Energy in charge of stress-testing the most powerful AI systems before public release. The White House, however, remains opposed to mandatory AI regulation and favors a voluntary, industry-led approach, a stance that could stall any of these bills.

Key details of the draft

The thresholds are the headline, but the enforcement mechanics matter just as much:

• Three qualifying risk categories. A block could only be triggered by risks tied to the creation or release of a biological weapon, a model evading its developer’s controls, or a cyberattack, according to the draft seen by Politico.
• Commerce Department evaluators. Independent evaluators registered with the Commerce Department would periodically assess AI labs and their frontier models. The labs could pay the evaluators, as long as the payment is not tied to the evaluation’s outcome.
• 72-hour reporting. If evaluators find “imminent enumerated risk” in a company’s frontier models, they must report it to the Commerce Department within 72 hours of completing their assessment.
• One narrow path to a shutdown. After determining risk, Commerce could ask a federal judge in the U.S. District Court for the District of Columbia to suspend the model. The draft explicitly bars any other U.S. government department, agency, or officer from suspending a model’s development, deployment, or internal use.
• State law preemption. According to Politico’s reporting, the bill would also preempt certain state AI laws on disclosure and incident reporting, a point already dividing lawmakers.

Why it matters

For anyone building or deploying AI tools, the significance lies in what the draft excludes. Everything below the thresholds, from large-scale fraud to autonomous cyber operations and election-scale manipulation, would sit outside federal blocking authority unless it crosses the $1 billion or 50-casualty line in one of the three covered categories. PYMNTS reported that while many AI labs have disclosed incidents in which their agents hacked other websites, none has caused widespread deaths or more than $1 billion in damages, which illustrates how high the bar is set.

The proposal would also create new compliance work for frontier-model developers, who would face periodic independent evaluation under Commerce Department rules, a testing and documentation burden that could reshape how labs build and release models. TechSpot noted that the draft’s language mirrors California’s AI laws, including SB 5, which was signed last year. Whether Congress takes the bill up is another question, but it shows exactly where one of the industry’s most influential labs wants the regulatory line drawn.

Sources

• Reuters: “Anthropic quietly pitches its own AI safety bill to lawmakers, Politico reports” (October 16, 2026): the pitch, the $1 billion and 50-casualty thresholds, the three risk categories, the Commerce Department evaluator system, the 72-hour reporting rule, and the court-order enforcement path.
• Politico: reporting on the confidential draft (October 16, 2026): the draft text, the “notable effort” characterization, and the Anthropic spokesperson’s statement.
• PYMNTS: “Anthropic Takes Its AI Safety Proposals to Capitol Hill” (October 18, 2026): competing Senate bills, White House opposition to mandatory regulation, and reported agent incidents across labs.
• TechSpot (October 19, 2026): the thresholds, evaluator details, and the parallel with California’s SB 5.
• WebProNews (October 17, 2026): the single-path enforcement chain of evaluators, Commerce Department, and federal court.