USA draft US AI bill said to come from Anthropic would let a court halt a model for 45 days

A draft federal AI bill reportedly linked to Anthropic is circulating online.

Per PPC Land on Oct 20, the 37-page document marked privileged and confidential would make the Commerce Department the regulator of the most powerful models, require outside evaluators every 180 days, and let a single federal court in Washington suspend a model judged close to causing mass harm for an initial 45 days.

It covers models trained with more than 10^25 operations, ten times below California's SB 53 line, with heavier duties for companies with over $500 million in AI revenue or $1 billion in development spending. It would also stop states from writing their own safety disclosure, incident reporting and evaluation rules.

Anthropic's name appears nowhere in the text; the link comes from a LinkedIn post citing Politico Pro, and PPC Land said it could not confirm the author. On Threads in Korea, critics said the bar for halting a model, an imminent mass-casualty event, amounts to "don't touch our AI."