What happened in this Tech story?

U.S. intelligence and cybersecurity agencies accused six China-based AI companies — DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI — of conducting industrial-scale distillation attacks to extract proprietary capabilities from U.S. frontier AI models including Claude, GPT, Gemini, and Grok.

What is a key fact in this story? (1)

**NSA, CISA, and FBI** jointly accused six China-based AI companies of running "industrial-scale" distillation attacks on U.S. frontier models since late 2024, extracting billions of tokens across millions of exchanges with likely Chinese government backing.

What is a key fact in this story? (2)

**DeepSeek** ran organized campaigns from late 2024 to mid-2025 to train its R1 and V3 models; **Moonshot AI** extracted Claude Fable 5 data for Kimi-K3 and GPT-4o data for Kimi-K2.

US Accuses Chinese AI Firms of Mass Model Distillation — SkimNews

By The Hacker News · Summarized & edited by Kapil Samant · 2026-09-09

Get the Tech newsletter

Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.

NSA, CISA, and FBI jointly accused six China-based AI companies of running "industrial-scale" distillation attacks on U.S. frontier models since late 2024, extracting billions of tokens across millions of exchanges with likely Chinese government backing.
DeepSeek ran organized campaigns from late 2024 to mid-2025 to train its R1 and V3 models; Moonshot AI extracted Claude Fable 5 data for Kimi-K3 and GPT-4o data for Kimi-K2.
Alibaba distilled Claude-4, Claude Opus, Claude Sonnet, and GPT-5 in late 2025 to improve software engineering, customer service dialogue, image/character creation, and RL/SFT/distillation integration in its own models.
MiniMax pulled CoT reasoning and software engineering capabilities from Claude Code, Claude Sonnet 4, Claude Opus, Gemini 1, Gemini 2.5 Pro, and Gemini 3 Pro to improve its M2 model; StepFun targeted Claude Opus 4.1/4.5, Sonnet 4.5, Haiku 4.5, and multiple GPT-5 variants for coding and agentic functions in Step 4.
Z.AI distilled billions of tokens of GPT-5.5 and Claude Opus 4.8 data as of mid-2026 to develop chain-of-thought reasoning capabilities.
Chinese firms achieved cost savings through bulk procurement of premium U.S. AI subscriptions shared across developer teams, routing requests through APIs, remote cloud providers, third-party aggregators, VPNs, obfuscated accounts, and proxy services marketed on Taobao and Xianyu.
Anthropic flagged DeepSeek, Moonshot AI, and MiniMax for similar campaigns back in February; Google Threat Intelligence Group separately reported a recent spike exceeding 100 million prompts targeting Google's visual, audio, image, and video generation capabilities.

Why it matters: This bulletin formalizes individual vendor complaints — Anthropic in February, Google this week — into a coordinated U.S. government position, giving frontier AI companies political cover to implement stricter detection, response manipulation, and cross-provider correlation. The named tactics, including bulk subscription purchases and proxy markets on Taobao and Xianyu, translate a terms-of-service violation into a documented national security threat with specific companies and model versions attached.