Anthropic, the company behind the AI 'Claude', has publicly condemned Chinese AI research institutions such as DeepSeek for allegedly using its technology without permission to enhance their own models. The company reported that these labs, including Moonshot and MiniMax, conducted a large-scale campaign to improve their models by illegally extracting functionality from Claude, utilizing around 24,000 fraudulent accounts to generate more than 16 million interactions with Claude, of which DeepSeek accounted for 150,000, Moonshot for 3.4 million, and MiniMax for 13 million.

The method employed by these Chinese research institutions involves a technique known as 'distillation'. This approach allows a lesser-performing model to be trained based on the output of an already developed model, significantly reducing the time and cost associated with AI model training. While it is permissible to use distillation for training one's own models, using another company's model without permission violates usage agreements. Anthropic has deemed this practice a national security issue, warning that foreign labs could supply authoritarian governments with capabilities derived from American models, thus undercutting American export controls intended to protect technological advantages.

In response to this challenge, Anthropic is reportedly investing in defensive measures to make it harder for competitors to execute distillation attacks and easier to identify such attempts. Despite these efforts, they stressed that the issue is not something any single company can resolve alone, highlighting the need for collaborative action from the AI industry and policymakers. However, Anthropic itself is facing scrutiny over its own model training practices, having been involved in multiple lawsuits for allegedly using copyrighted internet data without permission, including a ruling last September that resulted in a $1.5 billion settlement with copyright holders for the illegal use of millions of books.