Overview
- Politico first reported Oct. 16 that Anthropic quietly circulated a confidential draft to lawmakers that would define how and when Washington may stop so‑called ‘frontier’ AI models.
- The draft bars government action except for narrowly defined catastrophic risks that foreseeably cause more than $1 billion in damage or kill or seriously injure over 50 people and ties those risks to biological weapons, loss of developer control, or cyberattacks.
- It would make a developer ‘frontier’ only if a model exceeded a compute threshold of 10^25 FLOPs and the company met a financial test such as more than $500 million in AI revenue or over $1 billion in AI R&D spending.
- The bill requires registered independent evaluators to run periodic tests, report any identified ‘imminent enumerated risk’ to the Commerce Department within 72 hours, and lets Commerce ask a D.C. federal judge to order a suspension while other agencies would be barred from ordering halts.
- Critics say the fixed numeric lines and narrow risk list could leave many harms unaddressed and favor large incumbents, while supporters say the approach creates predictable rules and could replace a patchwork of state laws.