⚖️Anthropic Drafts AI Safety Bill That Can Halt Risky Models
TL;DR
Anthropic has circulated a draft federal bill that could block a model's release if over 50 people could die or $1B in damage is foreseeable. Evaluators would report to Commerce within 72 hours.
Anthropic has circulated a draft federal bill that could block a model's release if over 50 people could die or $1B in damage is foreseeable. Evaluators would report to Commerce within 72 hours. It goes further than most labs have been willing to back.
Key Points
Trigger: foreseeable, imminent risk of 50+ deaths or $1B+ in property damage
Qualifying risks: bioweapons, a model evading developer control, or cyberattacks
Independent evaluators register with the Commerce Department; labs can pay them if pay is not tied to results
Evaluators have 72 hours to notify Commerce, which then asks a D.C. judge to halt release
Thresholds mirror California's SB 53; federal agencies could not halt distribution on their own
Why It Matters
A leading lab is asking for enforceable pre-release checks while others push voluntary ones. How evaluators stay independent when labs pay them is still unanswered.
Quick Facts
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,569 builders reading daily.