Anthropic has publicly accused three Chinese AI firms—DeepSeek, Moonshot AI, and MiniMax—of using more than 24,000 fake accounts to exploit its Claude AI model. The companies reportedly generated over 16 million exchanges with Claude, leveraging a method called "distillation" to siphon advanced capabilities from the model.
The accusation centers on these firms targeting Claude's highly differentiated features, such as agentic reasoning, tool use, and coding, to enhance their own AI offerings. Anthropic said MiniMax redirected almost half its traffic to siphon off these capabilities, demonstrating systematic misuse.
This controversy comes amid an ongoing US debate over tightening export controls on advanced AI chips intended to limit China’s access to cutting-edge AI technology. The alleged abuse of Claude highlights vulnerabilities in current AI model protections and the challenges of safeguarding intellectual property in AI development.
Distillation, while a common AI training technique for creating smaller, efficient models, can be exploited by competitors to effectively copy others' technology. Similar allegations have been raised by OpenAI against DeepSeek, suggesting a broader pattern of Chinese AI labs leveraging foreign models to accelerate development.
The unfolding dispute underscores risks for AI companies over unauthorized use of models, which could undermine innovation and market competition. It also stresses the strategic importance of US policy decisions on AI chip exports and protecting AI intellectual property.
Industry observers will closely watch how regulators respond to these accusations and whether export controls will be tightened, potentially reshaping the AI competitive landscape between the US and China. Additionally, how companies adapt safeguards against distillation-based model theft could affect future AI development practices.