HR-10180-119
Referred to the House Committee on Science, Space, and Technology.
Sponsored by George Whitesides (D-CA)
What it does
This bill would direct the NIST Director to assess trends in AI systems' ability to autonomously conduct research or development on other AI systems. It would also require that any memorandum of understanding NIST signs with a frontier AI developer for pre-deployment evaluation include developer disclosure of how much AI was used to build the system and an assessment of the system's ability to self-improve without human oversight.
Who benefits
NIST, which gains explicit statutory authority and a clearer mandate to evaluate AI self-improvement risks. Federal policymakers and safety researchers who would gain access to standardized disclosure data from frontier AI developers that participate in these voluntary evaluations. The public indirectly, if earlier detection of autonomous AI R&D capabilities helps inform future regulation.
Who is hurt
Frontier AI developers who choose to enter pre-deployment evaluation agreements with NIST would face new disclosure obligations, including sharing proprietary metrics on AI-assisted development. Developers may also bear compliance costs in generating estimates and methodology descriptions requested by NIST, and some may view these disclosures as revealing competitively sensitive information.
Supporters argue
Supporters argue that as AI systems increasingly assist in their own development, regulators need visibility into this trend before autonomous AI research capabilities outpace oversight mechanisms. They contend that conditioning voluntary NIST evaluation agreements on disclosure requirements is a modest, low-cost way to build early-warning capacity without imposing mandatory rules on the broader AI industry.
Opponents argue
Opponents argue the bill only affects developers who voluntarily seek NIST evaluation agreements, so it may not reach the most secretive or highest-risk developers who decline to participate. They contend the vague terms "frontier artificial intelligence system" and self-improvement "capabilities" leave NIST broad discretion to define scope, creating uncertainty for developers without clear statutory limits.