Which Chinese AI companies were named in the advisory?

The advisory named six: DeepSeek, Moonshot AI, Alibaba Group, MiniMax, StepFun, and Z.AI.

Which U.S. models were targeted?

The targeted models included Anthropic's Claude, OpenAI's GPT, Google's Gemini, and xAI's Grok.

What is knowledge distillation?

It is a general machine learning technique, but the advisory says Chinese firms systematically abused it through proxy networks and account compromise to extract frontier model capabilities.

Large Language ModelsITmedia AI＋1h agoSep 17, 2026, 10:01 JST

NSA, CISA, FBI: Chinese firms siphoned Claude, GPT billions of tokens

Send on LINE

3 Key Points

What happened
The NSA, CISA, and FBI issued a joint advisory on September 8, 2026 saying Chinese firms including DeepSeek, Moonshot AI, and Alibaba extracted billions of tokens from U.S. frontier models like Claude and GPT.
Why it matters
The advisory points to large-scale distillation — copying a model's outputs to train a rival — as damaging to U.S. AI developers' research investments, and recommends countermeasures.
What to watch
The finding rests on tracing proxy infrastructure and compromised accounts, so its impact hinges on whether U.S. model providers tighten access controls and how Chinese firms respond.

WHO IT HITSU.S. frontier model providers like Anthropic, OpenAI, Google, and xAI — whose access controls and anti-distillation safeguards are the target of the recommended countermeasures — and enterprise security teams at companies evaluating or deploying Chinese AI models like DeepSeek's.

Not sure about something? Ask the AI

Summaries like this, in your inbox every morning.

Get news like this every morning — free

FAQ

Which Chinese AI companies were named in the advisory?: The advisory named six: DeepSeek, Moonshot AI, Alibaba Group, MiniMax, StepFun, and Z.AI.
Which U.S. models were targeted?: The targeted models included Anthropic's Claude, OpenAI's GPT, Google's Gemini, and xAI's Grok.
What is knowledge distillation?: It is a general machine learning technique, but the advisory says Chinese firms systematically abused it through proxy networks and account compromise to extract frontier model capabilities.

ITmedia AI＋Read Original Article

Send on LINE

Get the latest Large Language Models news every morning

For example, today's edition would include:

Anthropic folds Claude Cowork into Claude chatSiliconANGLE AI · 1h ago
Anthropic explains how Claude Code burns tokensITmedia AI＋ · 1h ago
Historian Andrew Scull talks psychiatry, ECT on Lex FridmanLex Fridman Podcast · 1h ago

AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.

Subscribe free →

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →