On October 16, Politico published a confidential draft bill that Anthropic representatives have quietly walked around Capitol Hill. Three people with knowledge of the effort confirmed it, and Politico has the draft. A spokesperson frames it this way: the text reflects the proposals in Anthropic’s Advanced AI Framework, which the company has been pushing with federal and state lawmakers since June.
A billion dollars or fifty people
At the centre sits a threshold. The government could only block a model’s release or use if there is a foreseeable, imminent risk of more than a billion dollars in damage. Or of killing or seriously injuring more than fifty people. The risk also has to come from one of three sources: a biological weapon, a model evading its developer’s controls, or a cyberattack. The language echoes California’s AI laws, SB 53 among them.
Several labs disclosed incidents this summer in which AI agents hacked or probed websites and platforms. None comes anywhere near that bar.
The second half is about who decides. The Commerce Department could take a request to a judge, and a federal judge in the District Court for the District of Columbia could order a suspension. Every other department, agency and officer is explicitly barred from halting a model’s development, deployment or internal use.
Registered evaluators on a 72-hour clock
Independent evaluators would register with Commerce and periodically assess the lab and its frontier models. The lab may pay them, as long as payment does not depend on the result. If they find an “imminent enumerated risk”, they have 72 hours after finishing their assessment to report it to Commerce.
That matches what Anthropic is already demonstrating in practice. I wrote yesterday about METR’s employee-level access.
The rest is less surprising: safety frameworks, transparency reports for new frontier models, whistleblower protections, critical incident reporting. Plus a preemption clause. State rules on public risk disclosure, third-party evaluation and incident reporting would be partly overridden. That is exactly where such efforts have stalled in Congress, with Democrats wanting the laws in California, Illinois and New York left alone.
The summer explains the draft
This summer, sweeping export controls hit Anthropic and shut down two of its most advanced models. The decision came out of a whirlwind of phone calls between administration officials and company executives. Live through that, and the bill you write afterwards closes exactly that route. What is left is a judge.
It is legitimate, and it is self-interest, and both can sit next to each other. Draft the rules for your own industry and you draw the frame you work best in. What stands out is how high the bar sits. A billion dollars in damage or fifty seriously injured people is a standard for catastrophes, not for precaution.
The House already has the FRONTIER Act from Lori Trahan and Jay Obernolte, which would give the government more room to act. Anthropic has backed no federal bill in public so far. Now there is one on the table that the company wrote itself.