Anthropic has provided U.S. lawmakers with a confidential federal draft bill on frontier AI safety that would allow the Department of Commerce to suspend a model facing a foreseeable and imminent risk of more than $1 billion in damages or the death or serious injury of more than 50 people. The proposal would limit federal intervention to three scenarios: assisting in the creation or release of a biological weapon, loss of control over a model, or a cyberattack. It would give shutdown authority to that single agency, require repeated independent evaluations and incident reporting, and draw on California’s SB 53 law and Anthropic’s Advanced AI Framework. Critics say the thresholds are too lenient and that an industry company should not write the rules governing its own sector, while Anthropic has committed to revising parts of the language by October 18, 2026. The draft has not been introduced in Congress and will compete with proposals such as the FRONTIER Act, leaving open the possibility of lower thresholds or broader coverage of risks.
Source: Read the original article
