Anthropic claims Chinese AI firms secretly use Claude. Is it true?

Chinese AI companies may have secretly utilized U.S.-based AI models like Anthropic's Claude to reduce costs, according to allegations from American AI lab Anthropic. The National Security Agency, FBI, and Cybersecurity and Infrastructure Security Agency recently warned that Chinese firms were engaging in large-scale "distillation" – a technique where the outputs from a more capable "teacher" model are used to improve another model.

Anthropic claimed that Chinese developers used fraudulent accounts and proxy networks to extract reasoning data from Claude, sometimes using the U.S. model as a concealed back end for their own applications. While model distillation is common in the AI industry, Anthropic alleged that Chinese labs bypassed usage restrictions through fraudulent accounts and 5,380 fraudulent accounts located primarily in Singapore and Japan.

Some Chinese companies reportedly used Claude as a hidden back end, with users unknowingly having their queries processed by the U.S. firm's servers. Industry experts are divided on the credibility and attribution of Anthropic's findings, with some considering them credible and others arguing that attributing direct blame to Chinese AI companies is not always straightforward.

Chinese regulators reportedly investigated domestic AI labs following Anthropic's report, focusing on DeepSeek and Moonshot, due to concerns over data security and the potential for sensitive information to be shared with American AI model providers.