A joint FBI, NSA and CISA alert named DeepSeek, Moonshot, Alibaba and three peers for industrial-scale copying of U.S. frontier models, sharpening the enforcement stakes ahead of the leaders' meeting.
뒤로
US Accusation of 6 Chinese AI Firms Raises Sanction Risk
U.S. security agencies formally accused six Chinese AI developers of systematically mining American models, escalating a technology rivalry that now threatens sanctions and export controls before the Sept. 24 Trump-Xi summit.
"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 FBI, National Security Agency and Cybersecurity and Infrastructure Security Agency said in a joint alert published Tuesday.
The agencies named DeepSeek, Moonshot AI, Alibaba Group, MiniMax, StepFun and Z.AI, saying they have drawn information from Anthropic's Claude, OpenAI's ChatGPT, Google's Gemini and xAI's Grok "at an industrial scale" since at least 2024. Distillation lets a developer use a leading model to train a cheaper rival and is typically barred by terms of service, though it is widespread across the industry.
The accusation lands weeks before President Donald Trump hosts Xi Jinping in Washington and follows Treasury Secretary Scott Bessent's July warning that any Chinese venture found to have stolen intellectual property could face sanctions. U.S. lawmakers are weighing measures to penalize companies that copy American models.
The report details which American systems each Chinese company distilled and for what purpose, including improving math problem-solving and code review. It said the campaign came "likely with Chinese government awareness" but stopped short of alleging an intelligence role. Chinese embassy spokesperson Liu Chang rejected the findings as "a deliberate attack on China's development and progress in the AI industry."
The accusation is the most detailed in a dispute that has simmered for more than a year. OpenAI said last year that DeepSeek had "inappropriately" based its model on ChatGPT. Google said in February it was hit with about 100,000 attempts to clone Gemini. In June, Anthropic accused Alibaba of running thousands of fraudulent accounts to gain access to Claude.
White House science adviser Michael Kratsios has pressed the issue since April, when he circulated a memorandum to federal agencies warning that distillation let Chinese companies "freeload off American innovation" and bypass safety guardrails. In July he said Moonshot had distilled Anthropic's Fable model to build Kimi K3, a release that stoked concern about China narrowing the gap with U.S. leaders.
The stakes extend beyond the six named companies. Alibaba and Moonshot rank among China's most valuable AI developers, and their models compete directly with American products in global markets. A sanctions or export-control response could raise the cost of U.S. chips and cloud services that Chinese developers rely on, while tightening the competitive dynamic that has driven rapid model releases on both sides.
The agencies advised U.S. developers to verify paying subscribers, share intelligence on suspicious behavior and, in some cases, downgrade responses to queries believed to be malicious. Whether Washington follows with export controls or sanctions will hinge on the summit's outcome and congressional action, with Bessent's July warning showing Treasury's willingness to act.
This article is for informational purposes only and does not constitute investment advice.
[1] Is China Stealing American AI? Why 'Distillation' Has Washington Up in Arms[2] U.S. agencies say top Chinese AI companies systematically copied American models[3] Alibaba, DeepSeek have 'systematically' siphoned AI models, says US[4] US says Alibaba, DeepSeek have ‘systematically’ siphoned AI models[5] U.S. agencies say top Chinese AI companies systematically copied ...
