US artificial intelligence company Anthropic accused three Chinese firms – DeepSeek, Moonshot AI, and MiniMax – of systematically extracting capabilities from its Claude chatbot through a large-scale data harvesting operation, a practice the company alleges constitutes industrial-scale intellectual property theft. The accusations, made public Monday, follow similar charges leveled against Chinese companies by OpenAI, the creator of ChatGPT, earlier this month.
Anthropic alleges the Chinese companies employed a technique known as “distillation,” utilizing outputs generated by Claude to rapidly improve the performance of their own, less advanced AI models. According to Anthropic, the operation involved approximately 16 million interactions with the Claude model, facilitated by roughly 24,000 fraudulent accounts. The company contends this allowed the Chinese firms to circumvent the substantial costs and complexities of independent AI development, as well as US export controls designed to maintain American leadership in the field.
“These campaigns are growing in intensity and sophistication,” Anthropic stated. “The window to act is narrow.”
Even as distillation is a recognized technique within AI development – often used to create smaller, more efficient versions of existing models – Anthropic argues the scale and intent of these operations crossed a line. The company claims the illicitly obtained capabilities pose national security risks, suggesting that models built through such methods may lack the safety protocols and guardrails implemented in Claude to prevent misuse, including applications in areas like bioweapons development and cyberattacks.
OpenAI made similar accusations to US lawmakers earlier in February, asserting that Chinese companies were actively using distillation techniques in “ongoing efforts to free-ride on the capabilities developed by OpenAI and other US frontier labs,” according to reports.
Anthropic identified MiniMax as the entity conducting the largest operation, generating over 13 million exchanges with Claude. The company stated that all three firms focused heavily on areas where Claude excels, specifically coding, agentic reasoning, and tool use. To bypass Anthropic’s restrictions on commercial access from China, the labs reportedly utilized proxy services to manage the extensive network of fake accounts.
The accusations come after the release of a low-cost generative AI model from DeepSeek last year surprised industry observers with its performance, rivaling that of leading American chatbots like ChatGPT. This development prompted questions about the US’s continued dominance in the AI sector.
Anthropic is calling for a coordinated response from both industry and government to address the issue, asserting that no single company can effectively tackle the problem alone. The company has not specified what concrete actions it expects from policymakers or other AI developers.
