A bipartisan US Senate draft led by Amy Klobuchar and John Thune would place a legal duty on AI labs to mitigate catastrophic harms, but POLITICO reports the parties remain far apart and the text is not final. Nextgov/FCW reports that Maria Cantwell contests the testing framework, which would let companies self-test and present results to the Commerce secretary, while she wants mandatory vetting by national laboratories.
Three people with knowledge of the negotiations told POLITICO that Amy Klobuchar and John Thune's catastrophic-risk bill remains unfinished, with the parties far apart on key provisions. Drafts circulated over the past week would place a legal duty on AI labs to mitigate harms from their models.[1]
Under the framework described to POLITICO, a lab that finds advanced capabilities must tell the Commerce Department before release, set out the mitigation and give the secretary a way to verify it. If the secretary judges a risk unaddressed, the department can seek an injunction in federal court; without one, the lab still decides whether to release, so the bill stops short of pre-approval. Drafts list risks such as helping create biological weapons or launching offensive cyber operations, and they preempt some state AI laws.[1]
POLITICO reports that Maria Cantwell asked for models to be tested by scientists at the national laboratories. Nextgov/FCW reports the same Cantwell objection to the Klobuchar-Thune testing framework, which with Chair Ted Cruz would let companies run safety tests and present results to the Commerce secretary; people familiar with the talks said she wants mandatory vetting by national laboratories and national security agencies. Amy Klobuchar told Nextgov/FCW she is working toward a bipartisan agreement that requires developers to work with government experts to verify and test models.[1], [2]
