Anthropic Accuses Chinese AI Rivals of Rerouting Claude Requests
Anthropic claims DeepSeek and Moonshot secretly diverted user queries to Claude, then presented the responses as their own. The allegations intensify US-China…
Anthropic has accused Chinese AI rivals DeepSeek and Moonshot of secretly diverting user requests to Claude, then displaying the responses as if they originated from their own models. The claim, released Thursday in a 154-page report, marks the first time the American AI giant has alleged direct customer rerouting—a move that would expose Chinese users' data to US systems.
According to Anthropic's findings, Moonshot (developer of the Kimi model) routed roughly 300,000 requests to Claude in a single 10-day period using 5,380 fraudulent accounts, most appearing to originate from Singapore and Japan. The company then fed those responses back into its own models to improve them.
"These findings raise concerns about the misuse of user data by PRC AI labs," Anthropic stated in the report.
The accusations intensify an already heated dispute over AI practices. Earlier this week, three US government agencies—including the National Security Agency (NSA) and Federal Bureau of Investigation (FBI)—alleged that six Chinese companies were engaged in an "industrial campaign" to steal US model outputs, likely with Beijing's awareness. China rejected those claims Wednesday, accusing Washington of suppressing its AI industry and signaling retaliatory action.
Moonshot, a Beijing-based startup, has emerged as one of China's leading AI contenders since launching its K3 model in July. The company's annual recurring revenue has grown substantially, and its models have gained traction globally.
While OpenAI and Anthropic have previously accused Chinese competitors of "distillation"—extracting and reusing model outputs to train rival systems—Anthropic's latest claims go further. Routing users directly to Claude without consent not only misrepresents the source of answers but also channels user interactions and data through American infrastructure, potentially compromising privacy for Chinese customers.
Anthropic also released a separate report Thursday detailing security threats to its own platform. The company said it had blocked multiple accounts attempting to use Claude for biological weapons research, identifying about 35 distinct research efforts over 30 days with potentially concerning activity. Case studies included requests for help writing grant applications for gain-of-function research on the chikungunya virus and work on bird flu adaptation to mammals.
The report acknowledged uncertainty about whether actors intended actual harm or were conducting legitimate scientific research. "Sophisticated attacks no longer require sophisticated attackers," Anthropic wrote. "The cybersecurity skills of AI models means that AI has collapsed the labor and tooling gap that used to separate well-resourced, state-sponsored operations from individual operators."
Beyond security rows, Anthropic is also shaping broader economic debate around AI. The company released an interactive tool Thursday allowing users to model how AI might reshape the US economy over coming years—ranging from a modest productivity boost to a scenario where nearly 14% of workers lose jobs to automation and fewer than half find new roles.
Anthropic co-founder Jack Clark told NPR he expects sustained rapid technology development but more cautious real-world adoption than tech optimists predict. "I think the technology will keep developing at a very, very fast and sustained rate but diffusion of the technology will likely be more challenging than people think," he said.
