Anthropic is quietly pitching U.S. lawmakers its own unpublished frontier AI safety bill, first reported by Politico on Oct 16 and flagged by Cointelegraph and Wu Blockchain. The draft would let the government seek to halt a model release or use only for foreseeable and imminent risk of more than $1B in damage or 50+ deaths or serious injuries. That risk must involve bioweapons, cyberattacks or the model escaping developer control. It would require regular independent evaluations, with evaluators registered with Commerce and reporting imminent risk within 72 hours, per summaries here and here. Only the Commerce Department could ask a federal judge in D.C. to block a model, while other agencies would be barred from stopping one alone. The draft would also place some state AI safety rules under a federal framework, described as preempting laws including California's. The bill text itself is not public, so all details are Politico reporting as summarized secondhand.