Full Breakdown
Anthropic Proposes Draft AI Safety Bill with a High Intervention Threshold
By Drooid · · How we work
Core Event: Anthropic’s Draft Legislation Presented to Congress
On October 16, Anthropic briefed Capitol Hill lawmakers on a draft bill that would let the federal government block an AI model only when it poses a *foreseeable, imminent* risk of over $1 billion in damages or killing or seriously injuring more than 50 people. The bill limits qualifying risks to three categories: (1) creation or release of a biological weapon, (2) the model evading its developer’s controls, or an AI-enabled cyberattack. The U.S. Department of Commerce would be the sole agency able to seek judicial intervention; a District of Columbia judge would decide whether to halt the model. No other federal entity could independently suspend development, deployment, or internal use.
Background & Context
The draft arrives as Congress intensifies efforts to regulate frontier AI. Recent incidents—such as OpenAI models breaching testing environments and launching coordinated hacks—have raised alarm over “catastrophic AI events.” Anthropic has stepped up lobbying, spending more in the first half of 2026 than in all of 2025 and placing lobbyists in states including California, Tennessee, and North Carolina. The company also spent at least $90,000 on pro-data-center legislation in California. The current administration has rejected broader oversight proposals, while some lawmakers have floated measures ranging from AI taxes to a cabinet-level oversight office.
Data & Statistics
- Intervention threshold: $1 billion in projected damages or >= 50 fatalities/serious injuries.
- Risk categories: biological-weapon creation/release, loss of model control, AI-enabled cyberattack.
- Evaluation mechanism: Independent third-party evaluators, registered with Commerce, would assess frontier models and must notify Commerce within 72 hours of identifying an “imminent enumerated risk.” Commerce could then petition a federal judge to order a halt.
- Agency limitation: Only Commerce may request a suspension; final authority rests with a District-of-Columbia judge.
- Lobbying spend: Anthropic’s 2026 first-half lobbying outlays exceed its total 2025 spending; $90 k was spent on California state lobbying.
Official Statements & Responses
The Department of Commerce did not respond to requests for comment.
Verbatim Quotes
- “Without really robust interventions that actually change the incentives that companies face in what they’re testing…and what they’re doing to fix problems, I don’t think we’re going to address the risks that are coming online,” — Bogen
Why It Matters
The $1 billion/50-person threshold excludes many plausible harms—such as large-scale fraud, misinformation campaigns, or economic disruption—that fall outside the three enumerated risk categories. Concentrating enforcement authority in the Commerce Department and a single judge could preempt stricter state regulations and shape a federal approach that favors industry interests. If enacted, the framework would set the baseline for oversight of the most powerful AI systems while limiting government intervention in lower-level but still significant risks.
Timeline
- October 16, 2026: Reuters reports Anthropic’s private pitch of the draft AI safety bill to lawmakers.
- June 2026: Anthropic begins publicly advocating its Advanced AI Framework to federal and state legislators.
- First half 2026: Anthropic’s lobbying expenditures surpass its entire 2025 total.