FBI and NSA accuse six Chinese AI firms of copying U.S. models

Federal agencies detail Chinese AI copying allegations

Three U.S. intelligence and security agencies have issued their most specific accusation to date that Chinese artificial intelligence companies are systematically copying American AI models, naming six firms and detailing which U.S. systems they allegedly distilled.

In an alert published Tuesday, the FBI, National Security Agency, and Cybersecurity and Infrastructure Security Agency said the Chinese companies have mined cutting-edge models from Anthropic's Claude, OpenAI's ChatGPT, Google's Gemini, and xAI's Grok since 2024 to build their own systems. The report was first detailed by NBC News.

The six companies named in the alert are DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI. According to the agencies, the distillation campaign came "likely with Chinese government awareness," though the report stopped short of claiming direct involvement by Chinese intelligence services.

What the agencies found

The federal alert goes into granular detail about the alleged copying, including a list of specific capabilities that certain Chinese models allegedly distilled from particular American systems. These include "legal specialization optimization," "agentic functions," and "coach/assistant capabilities."

"The sheer scale of these campaigns and their sophistication indicate that distillation is not a supplement to these companies' AI model development, but the critical core of it," the report stated.

Distillation is a common practice in the AI industry where a smaller model learns from a larger one's outputs, though it often violates the terms of service of commercial AI systems. The technique allows developers to create more efficient models but can also enable competitors to replicate proprietary capabilities without investing in the underlying research.

Why it matters

The detailed public accusation escalates U.S.-China tensions over AI development ahead of a scheduled September 24 meeting between President Donald Trump and Chinese President Xi Jinping. It also provides concrete examples to support the White House's broader narrative that the U.S. is in a technology race with China, one where intellectual property theft threatens American innovation advantages. For AI companies, the alert underscores the challenge of protecting proprietary models while offering commercial access, and may prompt stricter verification and monitoring of API usage.

Recommendations and context

The agencies recommended that U.S. companies improve verification of users who subscribe to their models, share information about suspicious behavior with one another, and in some cases downgrade or alter responses to queries believed to be malicious.

Michael Kratsios, director of the White House Office of Science and Technology Policy, has been highlighting Chinese distillation as a critical issue since at least April, when he circulated a memorandum to federal agency heads. In July, Kratsios publicly accused Moonshot of distilling Anthropic's Fable model to develop its Kimi K3 system.

Scattered accusations have emerged since last year, when OpenAI said DeepSeek had "inappropriately" based its model on ChatGPT. In February, Google reported being inundated with attempts to clone Gemini but did not name suspects at the time.

A spokesperson for China's embassy in Washington did not immediately respond to a request for comment on the allegations.

The details were first reported by NBC News reporters Kevin Collier and Jared Perlo.