Overview
- Anthropic handed a confidential draft federal bill to Capitol Hill that was reported on Oct. 16, proposing a legal framework for when the government can stop advanced AI models.
- The draft sets high intervention thresholds that would allow government action only for foreseeable, imminent risks exceeding $1 billion in damages or causing more than 50 deaths or serious injuries.
- Those thresholds apply only to three specific danger zones: creation or release of biological weapons, a model escaping its developer’s controls, and cyberattacks.
- The bill would require independent evaluators registered with the Commerce Department to run periodic tests, forbid pay tied to outcomes, and force 72‑hour reporting to Commerce of any identified imminent risks.
- Only Commerce could ask a federal judge in the District of Columbia to halt a model while other agencies would be barred from suspending development; the draft builds on California’s SB 53 and could shape upcoming congressional debate and costs for AI labs.