Anthropic has accused three major Chinese AI companies—DeepSeek, MiniMax, and Moonshot AI—of engaging in large-scale, unauthorized use of its Claude AI system to enhance their own models through a technique known as distillation. In a detailed statement, Anthropic described the activity as an "industrial-scale campaign" that violated its terms of service and regional access restrictions. The company revealed that approximately 24,000 fake Claude accounts were created and used to generate over 16 million interactions with the AI system, all without authorization. Distillation involves training a smaller, less powerful model using the outputs of a more advanced model, a practice that is legitimate when done ethically and transparently. However, Anthropic argues that these Chinese labs are using the method illicitly to rapidly acquire capabilities that would otherwise take years and billions in investment to develop independently. Anthropic emphasized that the threat is not limited to a single company or region. It warned that the window for effective action is narrowing and called for urgent, coordinated efforts among industry leaders, governments, and the global AI community. The company cited specific examples of the scale and sophistication of the attacks. For instance, DeepSeek was found attempting to create versions of AI that could bypass censorship on sensitive topics. Meanwhile, MiniMax was observed shifting nearly half its traffic to target Anthropic’s latest model within 24 hours of its release, showing a high level of responsiveness and coordination. Despite the lack of immediate responses from the Chinese firms, Anthropic said it detected MiniMax’s campaign while it was still active, allowing it to study the tactics in real time. The company has implemented behavioral fingerprinting systems to identify and block such attacks and shares threat intelligence with other AI developers to strengthen collective defenses. Beyond competitive concerns, Anthropic raised alarms about security risks. Less-trained models developed through distillation may lack critical safety safeguards, potentially enabling harmful applications such as the design of bioweapons. CEO Dario Amodei has previously warned that advanced AI systems could soon provide dangerous capabilities if not properly controlled. The company also reiterated its support for U.S. export controls on advanced AI chips, arguing that limiting access to high-performance hardware in China would hinder both direct model training and the ability to conduct large-scale distillation attacks. This stance contrasts with some other tech leaders, including Nvidia’s Jensen Huang, who believe such restrictions are ineffective and may harm U.S. competitiveness. Anthropic has itself faced criticism over its training data practices. In January, reports surfaced about Project Panama, an initiative described as an effort to scan nearly all books in the world for training data. Last year, the company settled a $1.5 billion class-action lawsuit related to the use of copyrighted material without permission, though it did not admit fault.