TL;DR
Anthropic reps have quietly pitched Capitol Hill lawmakers a draft bill to regulate frontier AI safety, Politico reported Friday, citing three people familiar with the effort. The government could block a model's release or use only if it poses a foreseeable, imminent risk of more than $1 billion in damage or of killing or seriously injuring over 50 people — and independent evaluators would be on the hook to flag that risk within 72 hours.
What Anthropic is actually pitching
This isn't a lobbying memo — it's draft legislation. Under the bill as Politico describes it, independent evaluators would register with the Commerce Department and periodically evaluate each frontier lab and its models. Labs could pay the evaluators, as long as the money isn't tied to the outcome of the evaluation.
If an evaluator finds what the draft calls “imminent enumerated risk” — biological weapons, a model evading its developer's controls, or a cyberattack — it must report to Commerce within 72 hours of finishing its assessment. Commerce could then ask a federal judge in DC to halt the model. Everyone else in government is explicitly locked out: the draft bars any other department, agency, or officer from suspending development, deployment, or internal use.
The bar is sky-high — deliberately
A government block requires a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously injuring more than 50 people. That is a catastrophic-harm trigger, not a product-safety standard. Day-to-day misuse, bias, privacy failures, agent incidents like the ones labs have disclosed this month — none of that clears the bar.
That's the tell. Anthropic is drawing the line where regulation touches existential-scale harm and preempting anything stricter. A bill written by the lab it would regulate is the lab's insurance policy against a tougher Congress writing its own version.
Why this week
The timing is not subtle. The White House's Super Intelligence Task Force held its inaugural meeting Thursday, with mandatory incident reporting for AI companies now on the table, and China published draft export controls on AI models, agents, and computing clusters the same day. Two governments moved on AI risk in one week.
An Anthropic spokesperson told Reuters the draft “represents proposals outlined in Anthropic's ‘Advanced AI Framework’ released earlier this year,” and that it would give the government the “ability to block or deter dangerous deployments of the most powerful models” — plus requirements that developers test frontier models, publish their findings, and submit to independent evaluation.
What to watch
This is a pitch, not a bill — nobody has introduced it, and Anthropic and Commerce both declined early comment. But drafts like this have a way of becoming the skeleton of whatever eventually passes, because lawmakers drafting AI law have very few off-the-shelf models to borrow from.
Watch for two things: whether any member of Congress picks the draft up and introduces it, and whether the other labs — OpenAI, Google, Meta — rally around it or fight it. If the industry consolidates behind one self-written rulebook, this week was the week AI regulation stopped being hypothetical.