Anthropic is quietly shopping its own AI safety bill, and it sets a high bar for stopping a model

Anthropic has been pitching lawmakers on Capitol Hill on a draft bill to regulate frontier AI safety and security, Politico reported on Friday, Oct. 16, 2026, and an Anthropic spokesperson confirmed the effort to Reuters.

It's unusual for an AI lab to write the rules for its own off switch, and the details are the story. The bar for pulling a model is high — an imminent risk of 50-plus casualties or a billion dollars — and the switch sits in one place: Commerce, with a judge's sign-off, with every other agency told hands off. You can read that as sensible guardrails against overreach, or as a company that's being blacklisted by the Pentagon making sure the Pentagon can't do it again. Either way, a frontier lab just handed Congress a draft. That's a lobbying move as much as a safety one.

Anthropic representatives have pitched Capitol Hill lawmakers on a draft bill to regulate frontier AI safety and security, Politico reported on Oct. 16, 2026, citing three people with knowledge of the effort. A frontier model, here, is one of the most capable general-purpose AI systems. The report does not say the bill has been introduced, name a sponsor, or print a bill number.

Per Politico: the government 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 stem from the creation or release of a biological weapon, the model evading its developer's controls, or a cyberattack. Those thresholds are Politico's account of the draft.

Per Politico: independent evaluators would register with the Commerce Department and periodically evaluate labs and their frontier models. Labs could pay the evaluators as long as payment isn't tied to the outcome. If an evaluator finds an "imminent enumerated risk," it must report to Commerce within 72 hours of finishing its assessment. Those lines are Politico's.

Per Politico: Commerce could ask a federal judge in the U.S. District Court for the District of Columbia to suspend a model. The draft explicitly bars any other U.S. government department, agency or officer from suspending a model's development, deployment or internal use. Those lines are Politico's.

An Anthropic spokesperson told Reuters the bill reflects proposals in Anthropic's "Advanced AI Framework," published earlier this year, and would give the government the "ability to block or deter dangerous deployments of the most powerful models," while requiring developers to test frontier models, be transparent about the findings and get them independently evaluated. Those sentences are the spokesperson's, as Reuters prints them. The Commerce Department did not immediately respond to Reuters.

Anthropic's published Advanced AI Framework says government should have legal authority to block or deter dangerous deployments, with civil penalties tied to global annual revenue that escalate with repeat violations. It proposes the rules apply only to models trained with more than 10 to the 25th floating-point operations — FLOPs, a count of the math steps used in training — from companies with more than $500 million in AI-related revenue or more than $1 billion in AI research-and-development spending. Those lines are Anthropic's framework.

The "only Commerce, only via a judge" clause sits next to a fight Anthropic is already in. The Pentagon put the company on a supply-chain blacklist, and the U.S. Court of Appeals for the D.C. Circuit upheld that label in September. Critics, including Mother Jones, have called the 50-death and $1 billion threshold alarmingly high. That criticism is theirs.

In plain terms, Politico reported on Friday that Anthropic has been shopping Capitol Hill a draft that would let the government stop a frontier model only if it posed an imminent risk of 50-plus casualties or a billion dollars in damage, and only if Commerce took that case to a D.C. federal judge. An Anthropic spokesperson confirmed the effort to Reuters and pointed at the company's Advanced AI Framework. No bill number has been reported.