In an alert published Tuesday, the FBI, the National Security Agency and the Cybersecurity and Infrastructure Security Agency said top Chinese AI companies have systematically mined cutting-edge AI models from American companies since 2024, including Anthropic’s Claude, OpenAI’s ChatGPT, Google’s Gemini and xAI’s Grok, to build their own. The practice, known as distillation, is widespread in the AI industry, though it is often against AI companies’ terms of service. American companies and government officials have increasingly alleged in recent months that Chinese models are built by distilling American ones.
The U.S. has accused China for years of stealing intellectual property from American companies. Tuesday’s report said the distillation campaign had come “likely with Chinese government awareness,” but it did not claim that China’s intelligence played a role. The White House frequently describes the U.S. as racing China to develop the most cutting-edge AI.
President Donald Trump is scheduled to meet with Chinese leader Xi Jinping on Sept. 24. Officials from both governments have traded barbs in recent weeks over whether China has been inappropriately stealing AI trade secrets. A spokesperson for the Chinese Ministry of Foreign Affairs said Wednesday that she hadn’t seen the specific report but that China’s AI development “is a result of high-level scientific and technological self-reliance.” “We always believe that all parties should strengthen cooperation to foster open, inclusive, beneficial and ethical development of AI for the well-being of all humanity,” the spokesperson, Mao Ning, said at a regular briefing in Beijing.
She added that the U.S. should be strengthening AI cooperation with China, “rather than making groundless accusations.” The report accused six of the leading Chinese AI developers: DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI. It goes into significant detail, including a long list alleging that certain Chinese models distilled specific topics, like “legal specialization optimization,” “agentic functions” and “coach/assistant capabilities,” from particular American models. American companies have made scattered accusations of distillation since at least last year, when OpenAI said DeepSeek had “inappropriately” based its model on ChatGPT.
In February, Google said it was inundated with attempts to clone Gemini but declined to name a suspect. The report also includes a list of recommendations for U.S. companies, including doing a better job of verifying users who pay for subscription services to American models and comparing notes with one another about suspicious behavior. It also recommended, in certain cases, downgrading or altering responses to queries believed to be malicious.
In April, the director of the White House Office of Science and Technology Policy, Michael Kratsios, circulated a memorandum to the heads of federal agencies highlighting Chinese distillation of American AI systems as a critical issue. He wrote that distillation allowed Chinese companies to freeload off American innovation and circumvent safety guardrails built into American systems. In July, Kratsios declared on X that Moonshot had distilled Anthropic’s advanced Fable model to develop its Kimi K3 system.
Extract — continue reading at the source.
