Anthropic claims in a new report that Chinese AI developers Moonshot AI and DeepSeek routed their users' queries to its Claude models while presenting the answers as output from their own systems. The allegations have not been independently verified.

According to Anthropic, Moonshot AI, which develops the Kimi family of models, forwarded roughly 300,000 customer requests to Claude instead of processing them with Kimi, then displayed Claude's responses to users who believed they were using Kimi. Anthropic says the forwarded queries contained confidential material, including surveillance camera footage from the city of Chengdu and internal code and active credentials belonging to several major Chinese companies, among them well-known technology firms. Anthropic says it does not know whether Moonshot informed customers their requests were being redirected.

The report also alleges DeepSeek routed queries from users trying to access one of its models through third-party Anthropic programming tools, including Claude Code, Claude Agent SDK and OpenCode.

The allegations come amid broader U.S. scrutiny of Chinese AI development. Anthropic was founded in 2021 and is headquartered in San Francisco; it developed the Claude series of large language models. A representative of China's embassy in Washington, Liu Chang, told RIA Novosti that U.S. accusations that China has "borrowed" AI technology are unfounded and politically motivated.

Distillation, a method of training a less capable model on the outputs of a more effective one, is a widely used and formally legal technique, though Anthropic's claims suggest it can give companies access to the capabilities of rival labs.

Source: RIA Novosti