Moonshot, DeepSeek secretly routed user requests to Claude, Anthropic claims

Anthropic, a prominent American AI company, has accused two Chinese rivals, DeepSeek and Moonshot, of surreptitiously directing user queries to Anthropic's Claude models, subsequently presenting those responses as originating from their own models. This revelation escalates the ongoing US criticism of China's AI industry practices.

Earlier this week, three US government agencies expressed concerns over six Chinese firms allegedly stealing US model outputs, possibly with Chinese government knowledge. Beijing dismissed these claims as Washington's attempt to stifle China's AI sector and warned of retaliation. Anthropic's accusations are unprecedented, as they are the first to publicly claim that Chinese firms direct users directly to Claude, thus exposing Chinese users' information to the American company.

According to Anthropic's 154-page report, Moonshot allegedly routed 300,000 requests through 5,380 fraudulent accounts, primarily located in Singapore and Japan, during a 10-day period. The company then passed these requests off as originating from Moonshot and utilized them to enhance their own models. This diversion allowed Anthropic to access sensitive data from Moonshot's customers in the US, Europe, and China, including potentially sensitive information from a user associated with China's People's Liberation Army.

Moonshot's models have gained significant traction globally, with its annual recurring revenue projected to reach $1 billion by the end of the year, up from $500 million in mid-2026.

DeepSeek also faces accusations of routing user requests to Claude, although the extent of its operation remains unclear. Anthropic accused five other Chinese companies of distilling their Claude models since February 2026, including Alibaba, Z.ai, Xiaomi, and MiniMax. Alibaba, owner of the South China Morning Post, did not respond to requests for comment. The accused companies have not provided a response to Anthropic's allegations.

Distillation, a common practice in AI development, involves using the outputs of larger models to train smaller ones. However, Anthropic argues that Chinese firms are engaging in "illicit" distillation on an industrial scale, bypassing Anthropic's restrictions on using their models in China through fake accounts and stolen credit cards.

Legal experts have noted potential violations of both US and Chinese laws. Joe Khawam, a US Department of State lawyer, highlighted the report's implications for data privacy and the timing of the revelations as a possible signal that the Trump administration will address the issue of Chinese companies' distillation practices during upcoming US-China AI talks.