S-3336-119
Read twice and referred to the Committee on Homeland Security and Governmental Affairs.
Sponsored by Margaret Hassan (D-NH)
What it does
The bill would require the Secretary of Homeland Security to start at least one prize competition on AI interpretability and at least one on AI adversarial robustness within 270 days of enactment, using existing prize authority in the Stevenson-Wydler Act. The Secretary would have to consult other federal officials and industry experts, and may use contracts or agreements with for-profit, nonprofit, state, local, territorial, or Tribal entities to run the contests. The Secretary would report to Congress within 180 days after the first competition ends. The bill authorizes $10 million for fiscal years 2026 through 2030; an authorization does not itself provide the money.
Who benefits
AI researchers, universities, startups, and nonprofits that could win prizes or contract work. Federal agencies and critical-infrastructure operators that may gain more understandable and attack-resistant AI tools. Congress, which would receive a report on research gaps. Users of AI systems in high-risk settings could see indirect benefits if the research produces usable methods or standards.
Who is hurt
Taxpayers would bear up to $10 million in authorized costs, subject to appropriation. DHS would take on administrative work that may divert staff and attention from other missions. Researchers and firms that compete but do not win would bear their own costs. Competing federal AI research efforts, such as those at NIST or NSF, might face overlap or compete for funds. If contest results are limited, the spending may yield little practical benefit.
Supporters argue
Supporters argue that AI systems are being used in high-stakes settings while their internal workings remain poorly understood and they stay open to attack. They contend that prize competitions are a low-cost way to draw in a wide pool of researchers, that $10 million over five years is modest, and that required consultation with NIST, NSF, and industry experts plus a final report give Congress useful information.
Opponents argue
Opponents argue that DHS is not the main federal science or standards agency for AI, and that NIST and NSF already fund related work, so the contests may duplicate it. They contend that prize contests may not solve hard basic-research problems like interpretability, that the broad definitions give DHS wide discretion over design, and that even a modest authorization adds spending without proof of results.
Constitutional context
Congress's spending power (Art. I, §8, cl. 1) and Necessary and Proper Clause support federal research funding and prize authority. The bill regulates no private conduct and restricts no speech, so it raises no significant constitutional question.
Checks and balances
The executive branch (DHS) gains a defined research-contest mandate with discretion over design, while Congress checks it through the authorization limit, the appropriations process, the required consultation, and the mandated report.
Historical precedent
Congress has previously authorized agency prize competitions under section 24 of the Stevenson-Wydler Technology Innovation Act, as amended by the America COMPETES Reauthorization Act of 2010, which this bill relies on.