Anthropic recently publicly accused Chinese AI companies DeepSeek, MiniMax and Moonshot of massively abusing its large model Claude to improve their product capabilities, defining this behavior as "industrial-scale abuse." The activity involved about 24,000 fake accounts and more than 16 million interactions with Claude, the company said in a statement on Monday, following a report by The Wall Street Journal.
According to reports, these companies are accused of using so-called "distillation" technology to extract capabilities from more advanced models like Claude to train smaller models. Anthropic admitted in its description that model distillation itself is a "legitimate training method", but also emphasized that this method "may also be used for improper purposes," such as obtaining powerful capabilities from other laboratories at a cost far lower than the time and capital cost of independent research and development.
Anthropic specifically warns that models obtained through illegal distillation in this way will most likely not inherit the security protection and restriction mechanisms in the original model. "Foreign labs distilling U.S. models can bypass these security constraints and feed unprotected capabilities directly into military, intelligence, and surveillance systems, enabling authoritarian governments to use cutting-edge AI for offensive cyber operations, disinformation proliferation, and mass surveillance," the company noted.
In this incident, DeepSeek is the one that has attracted the most attention - this company has caused shock in the global AI industry with its more computationally efficient and powerful models. Anthropic said there were more than 150,000 interactions between DeepSeek and Claude, with a particular focus on Claude's reasoning abilities. In addition, DeepSeek is also accused of using Claude to generate "censorship-friendly alternative expressions" to answer politically sensitive questions involving dissidents, party leaders, or authoritarianism to circumvent the filtering of the country's censorship system.
Doubts about DeepSeek come not only from Anthropic. Just a week ago, in a letter to U.S. lawmakers, OpenAI also accused DeepSeek of using similar distillation methods to "free ride" on the model capabilities developed by OpenAI and other U.S. companies for a long time to seek competitive advantage. The letter also positions DeepSeek as a threat to the U.S. AI ecosystem and technological leadership.
At present, Anthropic has not disclosed in detail the subsequent legal or regulatory action path in a public statement, but by raising the matter to the level of national security and global AI governance, the company clearly hopes to push regulators and policymakers to face the risk of "model distillation being abused." At a time when the global AI competition is becoming increasingly fierce and cross-border technology flows are highly sensitive, this accusation has further exacerbated the tension surrounding the competition and security game between China and the United States in AI technology.