Bill > S3336

This bill establishes two prize competitions to be administered by the Secretary of Homeland Security focused on advancing artificial intelligence (AI) research in two critical areas: interpretability and adversarial robustness. The first competition aims to improve how humans can understand AI decision-making processes, seeking research that makes AI models more transparent and comprehensible, with participants submitting interpretability frameworks and models across potentially multiple competition phases. The second competition targets developing AI systems that can resist malicious attempts to manipulate or compromise their functionality, with a focus on high-risk, high-impact applications. Both competitions require consultation with various government agencies and industry experts, and will evaluate submissions based on their potential to create broadly applicable principles and practical value in improving AI reliability and security. After the competitions conclude, the Secretary must submit a report to congressional committees assessing the results, identifying research gaps, and recommending potential future actions. The bill defines key terms like "adversarial robustness" (AI's ability to resist attacks) and "interpretability" (how well humans can understand AI decision-making), and authorizes $10 million in funding for these initiatives across fiscal years 2026-2030. The overall goal is to enhance the scientific understanding and technological resilience of artificial intelligence systems.

Military Affairs and Security

https://www.congress.gov/bill/119th-congress/senate-bill/3336/all-info