Anthropic representatives have pitched U.S. lawmakers a draft AI safety bill that would restrict when the government could block an advanced model’s release or use, POLITICO reported on Oct. 16. The outlet cited three people familiar with the effort and a confidential draft it obtained.
Under the reported proposal, intervention would require a foreseeable, imminent risk of more than $1 billion in damage, or of killing or seriously injuring more than 50 people. The risk would have to involve a biological weapon’s creation or release, a model escaping its developer’s controls, or a cyberattack. The threshold concerns anticipated harm, not casualties that must already have occurred.
POLITICO describes a court-based process: Commerce could ask a judge to halt a model’s release, and a federal judge in Washington, D.C., could order a suspension. The draft would also override some state requirements covering risk disclosure, outside evaluations and critical-incident reporting.
An Oct. 18 post questioned whether one death or 50 injuries would be the threshold for unplugging an AI. That characterization differs from the report’s more-than-50 threshold for deaths or serious injuries and its focus on imminent risk.
