AIBRIEF
A group led by Thune, Cruz, and Klobuchar is finalizing legislation that would let Washington block release of advanced AI models and preempt state laws on certain risks.
A bipartisan group of U.S. senators is finalizing legislation that would create a "duty of care" (a legal obligation to design products that prevent foreseeable harms) for developers of the most advanced AI systems. The duty would target catastrophic risks such as the design of nuclear or biological weapons by a bad actor.
The proposal would let the federal government block the release of models it deems unsafe, with companies able to challenge in court. It would also preempt state laws on certain AI risks, concentrating catastrophic-risk rulemaking in Washington. Scope is narrow by design: "most advanced" models only, a category the proposal names as including Alphabet's Google, Anthropic, and OpenAI.
Senate Majority Leader John Thune, Commerce Chair Ted Cruz, and Senator Amy Klobuchar are driving the negotiations, with Senate Commerce ranking member Maria Cantwell also weighing in. Cruz has said he is working with Klobuchar and Thune on a duty of care focused on "catastrophic risks involving biological or nuclear threats," per a Reuters wire by Courtney Rozen. The remaining reporting comes from two anonymous Senate aides and one anonymous lobbyist.
The text is not yet public. The House has roughly one week of session and the Senate about three before the November 3 midterms.