US claims Chinese AI firms are carrying out ‘industrial-scale’ theft of trade secrets

Peter Catterall/AFP/Getty Images

Chinese artificial intelligence firms are committing “industrial-scale” theft of their American counterparts’ trade secrets to save money and bring the coveted technology to market faster, US federal agencies alleged Tuesday.

Major Chinese tech companies are using “distillation,” which involves training lesser AI models on more advanced ones, to “extract restricted proprietary functionalities and capabilities of U.S. frontier AI models,” the FBI, National Security Agency, and Cybersecurity and Infrastructure Security Agency claimed in an advisory.

Distillation can be a legitimate research technique, but American officials and AI executives have long complained that the volume and targeted manner of Chinese distillation campaigns against US firms amount to stealing. The new advisory puts more evidence behind those accusations and adds to the growing tensions between US and China over AI technology. President Donald Trump has said that AI will be among the discussion topics when he hosts Chinese leader Xi Jinping in Washington, DC, later this month

“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 new advisory says.

China-based AI firms “use a gray market of proxies known as ‘transfer stations’” to get around US AI firms’ geographic restrictions, safeguards, and terms of use, the US agencies said. They singled out several Chinese tech firms, including DeepSeek and Alibaba Group, as having conducted “high-volume knowledge distillation campaigns” against US AI firms.

Treasury Secretary Scott Bessent threatened sanctions against Chinese companies that distill American AI.

“When [People’s Republic of China] firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table,” he wrote on X.

CNN has requested comment from the Chinese Embassy in Washington on the US agencies’ accusations, as well as from DeepSeek and Alibaba.

Anthropic alleged that three Chinese AI labs had been behind an “industrial-scale campaign” to extract capabilities from Anthropic’s AI model, Claude. The Chinese labs “generated over 16 million exchanges with Claude through approximately 24,000 fraudulent accounts,” Anthropic claimed .

Chinese open-weight AI models are growing in popularity since they are highly customizable, can operate off the cloud and tend to be less expensive.

The global race for AI supremacy has put a bull’s-eye on American firms in other ways. Hackers of various stripes are “increasingly targeting AI assets,” Google’s Threat Intelligence Group said in a report released Tuesday.

Hackers with a range of motivations have targeted “proprietary AI models and source code,” the report said. “This shift underscores that enterprise AI assets—from model weights to cloud compute quotas—are high-value targets for espionage, extortion, and resource theft.”

CNN’s Hadas Gold contributed to this report.

Peter Catterall/AFP/Getty Images

美国联邦机构周二指控，中国人工智能公司正在“大规模”窃取美国同行的商业机密，以节省资金并更快地将这项备受追捧的技术推向市场。

美国联邦调查局、国家安全局和网络安全与基础设施安全局在一份咨询报告中声称，中国主要科技公司正在使用“蒸馏”技术，即利用较高级的人工智能模型训练较低级的人工智能模型，以“提取美国前沿人工智能模型的受限专有功能和能力”。

蒸馏技术本身可能是一种合法的研究方法，但美国官员和人工智能企业高管长期以来一直抱怨，中国针对美国公司开展的大规模、有针对性的蒸馏活动，无异于窃取知识产权。这份新的咨询报告为这些指控提供了更多证据，并加剧了美中两国在人工智能技术领域日益紧张的关系。美国总统唐纳德·特朗普表示，本月晚些时候，他将在华盛顿接待中国国家主席习近平，人工智能将是双方讨论的议题之一。

这份新的咨询报告指出：“这些活动的规模之大、复杂程度之高表明，提炼不是这些公司人工智能模型开发的补充，而是其关键核心。”

美国机构表示，总部位于中国的AI公司“利用被称为‘转运站’的灰色代理市场”来规避美国AI公司的地域限制、安全保障措施和使用条款。他们特别指出，包括DeepSeek和阿里巴巴集团在内的几家中国科技公司曾对美国AI公司开展“大规模知识蒸馏活动”。

美国财政部长斯科特·贝森特威胁要对盗用美国人工智能技术的中国公司实施制裁。

“当（中华人民共和国）公司进行秘密的、工业规模的蒸馏攻击，越过知识产权盗窃的界限时，制裁和实体清单的指定将会摆上台面，”他在X上写道。

CNN已就美国机构的指控向中国驻华盛顿大使馆、DeepSeek和阿里巴巴请求置评。

Anthropic 声称，三家中国人工智能实验室策划了一场“大规模的行动”，试图窃取 Anthropic 人工智能模型 Claude 的功能。Anthropic 还声称，这些中国实验室“通过大约 24,000 个欺诈账户与 Claude 进行了超过 1600 万次交互”。

中国开源人工智能模型因其高度可定制性、可在云端运行且价格往往较低，而越来越受欢迎。

全球人工智能霸权之争也以其他方式将美国公司置于聚光灯下。谷歌威胁情报小组周二发布的一份报告称，各种类型的黑客“正越来越多地将目标对准人工智能资产”。

报告指出，出于各种动机的黑客将目标对准了“专有人工智能模型和源代码”。“这种转变凸显了企业人工智能资产——从模型权重到云计算配额——都是间谍活动、勒索和资源窃取的高价值目标。”

CNN的哈达斯·戈尔德对本报道亦有贡献。