Bipartisan legislation aims to combat AI distillation attacks with federal guidance and industry collaboration

A bipartisan House bill, the AI Cyber Defense Act, seeks to establish federal guidelines for defending AI models against distillation attacks, with the National Institute of Standards and Technology (NIST) tasked to develop voluntary industry standards. The bill, sponsored by Reps. Pat Harrigan and George Whitesides, targets countries including Russia, China, Iran, and North Korea, which are accused of systematically copying U.S. AI models. Harrigan emphasized the high cost of maintaining a U.S. lead in AI research, noting that adversaries like China may strategically delay their own investments to steal U.S. advancements. NIST will also define distillation attacks, distinguishing them from legitimate model collaboration. The bill also proposes an AI Cybersecurity Roundtable at NIST to foster industry and government collaboration on cybersecurity standards for critical infrastructure. Rep. Whitesides highlighted the importance of slowing down adversarial progress to build necessary safeguards for AI technology. The legislation defines 'entities of concern' as foreign organizations based in the four targeted countries or operating under their direction. NIST’s director can also designate specific individuals or organizations as such. U.S. AI companies face lawsuits over alleged unauthorized use of creators’ works in training models. The bill aims to create a framework for defensive AI use, with input from industry experts and stakeholders, to protect critical infrastructure from cyber threats.