US security agencies have accused six Chinese companies of systematically exploiting American AI models to develop their own systems.
The agencies, including the National Security Agency and the Federal Bureau of Investigation, said the activity amounted to "aggressive, malicious, and targeted distillation activities at an industrial scale."
What distillation actually means
The technique, known as distillation, trains a new AI model on the outputs of a stronger system.
In practice, a smaller model queries a more advanced one repeatedly and learns from the responses it receives, effectively absorbing some of the original model's capabilities at a fraction of the cost.
Distillation is common practice across the AI industry when carried out with permission, but the agencies allege these efforts were unauthorised and conducted at scale to get around usage rules.
Who has been named
The joint advisory names DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI as the companies involved.
The agencies said the activity dates back to at least late 2024 and was carried out "likely with Chinese government awareness."
Officials said the six companies targeted models from Anthropic, OpenAI, Alphabet's Google and SpaceX's xAI, including various versions of Claude, ChatGPT, Gemini and Grok.
DeepSeek is separately accused of running organised distillation campaigns against several Claude models to help build its R1 and V3 systems, while Moonshot AI is accused of extracting data from Claude to help train its Kimi K3 model, released in July.
Kimi K3 is described in company filings as an open-weight model with around 2.8 trillion parameters.
A pattern building for months
The accusation builds on earlier disclosures, including an OpenAI memo to Congress that separately accused DeepSeek of "free-riding" on the capabilities developed by American frontier labs.
Anthropic has previously said three Chinese AI laboratories, including DeepSeek, Moonshot and MiniMax, generated more than 16 million exchanges with Claude through roughly 24,000 fraudulent accounts.
Timing that raises the stakes
The advisory lands weeks before President Trump is due to meet Chinese President Xi Jinping, adding a diplomatic edge to an already tense standoff over AI competition between the two countries.
Officials from both governments have traded accusations in recent weeks over whether China has been improperly extracting American AI trade secrets, a dispute this advisory is likely to intensify rather than resolve before the two leaders sit down together.
