Senate AI safety bill path uncertain amid language disputes

Drafts of the proposed legislation outline categories of risks that AI companies would need to consider, including the potential for models to create biological weapons or launch offensive cyber operations. According to sources familiar with the matter, AI labs would be required to disclose findings about models with advanced capabilities to the Commerce Department. This disclosure would also need to include information on the steps taken to counter identified risks and a method for the secretary to verify that these dangers have been addressed. Such disclosures would need to occur before a model is released, allowing the secretary time for review.

Under the proposed framework, if the secretary of Commerce determines that a particular risk has not been adequately mitigated, the department would have the authority to pursue an injunction in federal court. However, without such an injunction, AI labs would retain the discretion to release a model without undergoing a pre-approval process. The bill also includes provisions that would preempt certain state-level AI laws.

Senate Commerce Chair Ted Cruz (R-Texas) has been actively participating in the negotiations and has previously advocated for legislation to address potential catastrophic risks associated with AI.