Anthropic's draft AI bill would restrict shutdowns to extreme cases
Anthropic has quietly shared a proposed text with congressional offices, which would allow a federal judge to halt an advanced model only in dire circumstances.
By Standard Topic. AI-generated from the reporting listed under Coverage. Published . Updated .
This story is no longer being updated. It reflects reporting up to .
What happened
Anthropic has privately passed a draft bill that would set federal rules for frontier AI systems to members of Congress. The draft would let the government halt a frontier model only in extreme cases and only through a federal judge, with enforcement routed through Commerce.
Why it matters
The draft would set the conditions under which the government could stop a frontier AI model. Its narrow triggers and its departure from Amodei's summer positions shape the debate over federal AI rules.
Timeline
- Oct 18, 3:22 PM ET: Reports said Anthropic was pitching Congress its own AI safety bill, which would let the government halt a frontier model only in extreme cases and only through a federal judge.
- Oct 18, 4:15 PM ET: Politico obtained the confidential proposal and spoke with three people familiar with it, as Anthropic privately circulated the draft to lawmakers.
- Oct 19, 11:04 AM ET: Reports said the draft routes all enforcement through Commerce and requires 50 deaths or $1 billion in damages before a model shutdown can be sought.
- Oct 19, 6:48 PM ET: The Washington Post reported the draft is markedly different from what Anthropic CEO Dario Amodei advocated this summer.
Coverage
- AI & Tech Brief: Anthropic’s leaked bill (The Washington Post)
- Anthropic’s Proposed AI Safety Bill Only Kicks In at 50 Deaths or $1 Billion in Damages (Yahoo News)
- Anthropic floats bill letting judges halt risky AI models
- Anthropic pitches lawmakers AI safety bill with model-blocking powers
- Anthropic proposes federal AI law letting Washington block risky models
- Anthropic pitches Congress an AI safety bill as co-founder urges rules
More from Technology & AI
- China puts humanoid robots to work serving food and sorting parcels
- Newsom proposes sweeping AI safety rules for Washington
- Google releases Gemini 4 Neon AI model with million-token context
- Microsoft and Intel widen their cloud chip partnership
- Google DeepMind rolls out Genie 4 to select subscribers
- AI helped AMD shorten chip design work, OpenAI says