American artificial intelligence giant Anthropic accused its Chinese rivals DeepSeek and Moonshot on Thursday of covertly routing customer requests to its Claude models, then displaying the responses to users as if they originated from their own systems. According to Anthropic’s 154-page report published Thursday, Kimi model developer Moonshot secretly directed user prompts through Claude, utilized 5,380 fraudulent accounts, most of which appeared to be located in Singapore and Japan to bypass geographic restrictions inside China, and funneled nearly 300,000 requests primarily to the Opus AI model to train its own software.
Moonshot and DeepSeek Routing Allegations by Anthropic
Anthropic stated that Moonshot diverted user queries without notifying customers, routing the traffic away from its Kimi model and directly into Claude. According to Anthropic’s report, one specific instance revealed nearly 300,000 customer requests sent primarily to Anthropic’s Opus AI model. Because Anthropic does not allow its technology to be accessed from inside China, Moonshot relied on 5,380 fraudulent accounts, most of which appeared to be located in Singapore and Japan. Anthropic described this practice as distillation, meaning a technique where a developer leverages information from a superior AI to upgrade a less advanced network. The report asserts that DeepSeek and Xiaomi Corp. engaged in similar practices, raising concerns about the misuse of user data by PRC AI labs.
US Government Industrial Campaign Claims and Beijing’s Response
These company-level allegations follow a further escalation earlier in the week, when three US government agencies—including the National Security Agency (NSA) and the Federal Bureau of Investigation (FBI)—asserted that six Chinese companies were participating in an industrial campaign to steal American model outputs, “likely with Chinese government awareness”. Beijing rejected those claims on Wednesday, with the Chinese Commerce Ministry asserting that there is no factual or legal basis for accusations concerning industrial-scale distillation of US models by Chinese AI companies. The Chinese government characterized Washington’s actions as an attempt to suppress China’s AI industry and threatened retaliatory countermeasures.
Did you know?
Model distillation is a process where a developer uses outputs from a more powerful AI model to improve a less powerful system. While OpenAI and Anthropic have repeatedly accused their Chinese counterparts of “stealing” their outputs to improve their own models in a process known as “distillation”, this new set of accusations from Anthropic represents the inaugural instance of a Chinese firm being charged with directly redirecting users to Claude, a move that would risk exposing Chinese user activity and data to the American enterprise.
Industry Responses and Future Implications
Representatives for Moonshot, DeepSeek, and Xiaomi did not immediately respond to requests for comment outside normal business hours.
Frequently Asked Questions
What did Anthropic accuse Chinese AI labs of doing?
Anthropic accused Moonshot and DeepSeek of covertly routing user queries to Claude models and presenting the responses to users as their own, while using the data to train their domestic software.
How did Moonshot bypass geographic restrictions?
According to Anthropic’s report, Moonshot utilized 5,380 fraudulent accounts, most of which appeared to be located in Singapore and Japan, to access Claude, since Anthropic does not allow its technology to be accessed from inside China.
What was Beijing’s response to the allegations?
The Chinese Commerce Ministry rejected the accusations as a US effort to suppress China’s AI industry, stating there is no factual or legal basis for claims of industrial-scale distillation and threatening countermeasures.
Stay Updated on Global AI Developments
Explore more industry analysis and subscribe to our newsletter for the latest updates on cross-border technology policy and security.
Keep reading