Politico says Anthropic pitched lawmakers a draft frontier AI safety bill

On 16 October 2026, Reuters reported that Anthropic representatives have quietly pitched Capitol Hill lawmakers a draft bill to regulate frontier AI safety and security. The wire, as carried by Investing.com, attributes the report to Politico, which cited three people with knowledge of the efforts. Bloomberg Law’s lead, also 16 October and also citing Politico, says Anthropic PBC has presented its own draft bill to US lawmakers. No bill text is public. Every provision below is what Politico reported, per Reuters. None of it is a filed bill or a law.

Thresholds Politico reported

Politico reported, per Reuters, that the bill would let the government 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. Politico also reported, per Reuters, that the risk must stem from the creation or release of a biological weapon, the model evading its developer’s controls, or a cyberattack.

On evaluation, Politico reported, per Reuters, that independent evaluators would register with the Commerce Department and periodically evaluate the lab and its frontier models, and that labs could pay those evaluators if the payment is not tied to the evaluation’s outcome. In the same Reuters account of Politico’s report, evaluators that determine there is “imminent enumerated risk” must report it to Commerce within 72 hours of completing their assessment.

On enforcement, Politico reported, per Reuters, that Commerce could ask a judge to halt a model after determining risk, and that a federal judge for the District Court for the District of Columbia could order the suspension. Reuters also says Politico reported that the language explicitly prohibits any other US government department, agency, or officer from suspending development, deployment, or internal use.

Bloomberg Law’s lead, citing Politico, gives a shorter version: the government could block the release of a model judged to imminently risk killing or injuring more than 50 people or causing more than $1 billion in damages, with the biological-weapon case named as one trigger. That lead does not cover the 72-hour clock or the Commerce and court structure. Law Commentary and a French copy of the Reuters wire repeat the longer list and attribute it to Politico.

What the spokesperson told Reuters

An Anthropic spokesperson told Reuters that the bill represents proposals outlined in Anthropic’s “Advanced AI Framework,” released earlier this year. The spokesperson said the bill would grant the government the “ability to block or deter dangerous deployments of the most powerful models,” and that it also requires developers to test frontier AI models, be transparent about findings, and get them independently evaluated. Reuters says the Commerce Department did not immediately respond to a request for comment.

That paragraph is the on-record statement in the Reuters copy. The 50-person line, the damage threshold, the 72-hour clock, and the District of Columbia judge are what Politico reported, per Reuters. The spokesperson’s statement, as Reuters carries it, does not address those specific figures.

What the June framework says

Anthropic’s Advanced AI Framework is a public PDF dated June 2026. The policy page and the framework page describe testing, independent evaluators, disclosure of risk assessments and safety incidents, and government authority to block or deter dangerous deployments.

The PDF’s definition of catastrophic risk, which it says is in line with California SB 53, is a foreseeable and material risk of materially contributing to significant death, injury, or damage. It does not set Politico’s thresholds of more than 50 deaths or serious injuries, or more than $1 billion in damage. The $1 billion figure in the PDF is a coverage test: a developer is in scope if it earns more than $500 million in annual AI-derived revenue (adjusted for inflation, the PDF says) or spends more than $1 billion a year on AI research and development, and if it trains models past a compute threshold the PDF sets.

On incidents, the PDF would require a report of a “Critical Safety Incident” within 15 days of discovery, not within 72 hours. Its enumerated risks are biological weapons, offensive cyber operations, loss of control, and automated research and development. Its enforcement section contemplates court remedies. It does not say that only Commerce, through one federal judge in the District of Columbia, may halt a model, or that every other agency is barred.

Reuters describes a pitch of a draft, not an introduced bill. Civil investigative demands reported this week, and eight states’ lawsuits against OpenAI, are other proceedings. In the June PDF the incident clock is 15 days, and the $1 billion figure decides who is covered, not when a model may be stopped.