Overview
- A joint advisory from the NSA, CISA and the FBI published Tuesday said DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI ran coordinated distillation campaigns that extracted billions of tokens from U.S. frontier models since late 2024.
- The agencies describe distillation as using a powerful 'teacher' model’s outputs to train a smaller 'student' model and say the accused firms routed millions of queries through APIs, cloud providers, aggregators and proxies to hide activity.
- Officials warned the campaigns cut R&D costs for the Chinese firms by shortcutting model training, and said the stolen capabilities could be repurposed for military or cyber uses that threaten U.S. and allied security.
- The advisory urged immediate technical steps for U.S. developers, including behavioral detection, response changes and cross‑provider intelligence sharing, and U.S. policymakers are weighing export controls, sanctions or Entity List actions.
- Beijing rejected the claims as unfounded and politicized, a split that raises diplomatic friction ahead of planned mid‑September AI safety talks and a late‑September summit between Presidents Trump and Xi.