Anthropic, a San Francisco-based company, has accused three prominent Chinese AI laboratories, DeepSeek, Moonshot AI, and MiniMax, of using 24,000 fake accounts to extract capabilities from its Claude models. The company claims that these labs generated over 16 million exchanges with Claude through these fake accounts, violating Anthropic‘s terms of service and regional access restrictions.
The alleged campaigns, which Anthropic says are the most concrete evidence of a practice known as distillation, involved the use of a technique where a smaller, more efficient AI model is created by extracting knowledge from a larger, more powerful model. DeepSeek, Moonshot AI, and MiniMax allegedly used this technique to capture capabilities that took years and hundreds of millions of dollars to develop. Anthropic claims that DeepSeek conducted the most technically sophisticated operation, generating over 150,000 exchanges with Claude, while Moonshot AI and MiniMax generated over 3.4 million and 13 million exchanges, respectively.
Anthropic has framed this issue as a national security crisis, arguing that models built through illicit distillation lack necessary safeguards, creating significant national security risks. The company claims that these models can be used to develop bioweapons, carry out cyberattacks, or enable mass surveillance. Anthropic has called for coordinated action among industry players, policymakers, and the global AI community to address this issue. The company has also outlined a multipronged defensive response, including building classifiers and behavioral fingerprinting systems to identify distillation attack patterns and sharing technical indicators with other AI labs and authorities.
The implications of Anthropic‘s disclosure are significant, with potential repercussions for the AI industry, policymakers, and regulators. The company’s decision to frame this issue as a national security crisis rather than a legal dispute may reflect the difficult reality that intellectual property law offers limited recourse against distillation. As Anthropic CEO Dario Amodei has argued, export controls are crucial in preventing foreign labs from acquiring advanced AI capabilities, and the company’s disclosure may galvanize support for stricter export controls. The era of treating model access as a simple commercial transaction may be coming to an end, replaced by one in which API security is as strategically important as the model weights themselves.
Leave a Reply