US names six Chinese AI firms in industrial-scale model distillation

Ars Technica AI · rss · 2026-09-10

In a joint advisory, the NSA, CISA and FBI named DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI, alleging industrial-scale distillation attacks against US frontier models (variants of Claude, GPT, Gemini, Grok) since at least late 2024, likely with Chinese government awareness.

Agencies said such distillation gives China-based firms significantly shorter development timelines and billions in savings on training frontier models.

Original post →

More from Safety

Safety channel →