Congress funds AI safety research to make algorithms more trustworthy

What it does

This bill directs the Department of Homeland Security to run two prize competitions—one to advance AI interpretability (making AI decision-making understandable to humans) and one to develop AI models that resist adversarial attacks (hacking attempts that trick AI into wrong answers). Winners compete for federal prizes funded by $10 million over five years. The bill requires DHS to consult with tech companies, other agencies, and academic experts, and to report back to Congress on what the competitions reveal about gaps in AI safety research.

Why we flagged it

The bill's core mechanism is straightforward: it authorizes DHS to conduct two federally-funded prize competitions focused on AI interpretability and adversarial robustness. This is a direct public investment in AI safety research, not a regulatory carve-out or industry subsidy.

What the text implies

• Prize competitions may create intellectual property or research outputs that benefit private AI companies without explicit licensing restrictions, potentially allowing commercial firms to incorporate publicly-funded research into proprietary products.
• The bill's requirement to consult with 'any industry expert from the artificial intelligence sector' gives tech companies informal influence over competition design and evaluation criteria, potentially shaping research priorities toward industry-friendly outcomes.

The full analysis lists 3 implications of this text.

Who stands to gain

AI research institutions and universities; AI software and infrastructure companies (AMD, HPE, Cisco, Cognizant, Verisign as tech ecosystem ben; Consulting and systems integration firms