US HB3919

NSA Director, through the Artificial Intelligence Security Center, must develop an "AI Security Playbook" to defend covered AI technologies from theft by threat actors including nation-states and highly resourced actors

Playbook must identify vulnerabilities in advanced AI data centers and developers, catalog critical components (models, model weights, training insights) that would benefit adversaries if stolen, and outline strategies to detect, prevent, and respond to cyber threats

Director must engage with prominent AI developers and researchers through document reviews, expert interviews, roundtable discussions, and facility visits, and collaborate with federally funded research centers that have studied AI security

Initial progress report due within 90 days of enactment; final report with completed Playbook due within 270 days, including classified and unclassified versions plus a publicly available version

Covered AI technologies include advanced AI systems that could pose grave national security threats if stolen, such as those matching human expert performance in CBRN matters, cyber offense, model autonomy, persuasion, or self-improvement