What happened
Michael Kratsios, White House Office of Science and Technology Policy Director, issued a formal memo (NSTM-4) to federal agency heads on April 23, 2026, accusing primarily China-based actors of running coordinated campaigns using tens of thousands of proxy accounts to distill and replicate U.S. frontier AI models, enabling foreign actors to strip safety guardrails and release models matching U.S. capabilities at a fraction of the cost.
Why it matters
This is the first formal U.S. government accusation of systematic AI intellectual property theft via distillation attacks, marking a significant escalation in U.S.-China AI competition. The memo names DeepSeek, Moonshot AI, and MiniMax as perpetrators and commits the Trump administration to sharing threat intelligence with U.S. AI companies and developing defensive best practices, potentially reshaping how frontier labs protect model access and API abuse.
Action needed
Security teams should review API rate limiting, implement advanced distillation detection mechanisms, and prepare for federal guidance on industrial-scale abuse patterns. Organizations deploying frontier models should assess whether current guardrails can withstand systematic extraction attempts and evaluate third-party access controls.