Overview
- A Politico report on Oct. 16 revealed Anthropic has quietly shopped a confidential draft bill to lawmakers that it says reflects its Advanced AI Framework.
- The draft would let the government block a model only when it poses a foreseeable, imminent risk of more than $1 billion in damage or of killing or seriously injuring at least 50 people.
- Qualifying risks are limited to three scenarios: creating or releasing biological weapons, a model escaping its developer’s control, or a cyberattack.
- The bill would require independent evaluators registered with Commerce to do periodic tests and to report any identified imminent enumerated risk to Commerce within 72 hours, after which Commerce could ask a D.C. federal judge to suspend a model while other agencies would be barred from ordering suspensions.
- The draft also defines which labs count as “frontier” using compute and financial tests (including a 10^25 FLOPs training threshold), would preempt some state rules, has drawn criticism for high and narrow thresholds, and remains a company‑circulated proposal that Congress has not adopted.