Article 287: Anthropic Drafts Its Own Frontier AI Safety Bill - With a $1 Billion Trigger

Anthropic has circulated a draft federal bill on Capitol Hill that would let the Commerce Department ask a federal judge to halt a frontier model, but only where the risk is imminent and projected harm exceeds $1 billion in damages or 50 or more deaths and serious injuries, across bioweapons, loss of control and cyberattack pathways. Developers would have to report qualifying incidents within 72 hours. It lands alongside the existing FRONTIER Act, and follows the same week in which the FTC subpoenaed both OpenAI and Anthropic over safety records.

Key Takeaway: The thresholds are the entire argument, and Anthropic chose them. A lab writing the law that governs it is not automatically a conflict of interest - an industry that understands failure modes better than legislators is worth listening to - but the specific numbers are a judgment call dressed as a technicality. A halt power that only fires after a billion dollars of damage cannot prevent the event it exists for; it can only punish it. The 72-hour reporting duty is the more operationally meaningful clause, because it forces disclosure regardless of whether the threshold is ever crossed.

Sources: Politico, Reuters, AIToolsRecap, 17 October 2026