Anthropic is once again in the news, publicly accusing three prominent Chinese AI companies — DeepSeek, Moonshot AI, and MiniMax — of running coordinated, large-scale operations to illegally extract and replicate its technology. In a detailed blog post, Anthropic claimed the firms created approximately 24,000 fraudulent accounts to generate over 16 million exchanges with Claude, employing a technique known as “distillation” to train their own competing models.
Distillation involves using outputs from a more advanced model to rapidly improve a less capable one, allowing competitors to acquire sophisticated capabilities, such as agentic reasoning, coding, tool use, and orchestration, without independent development. Anthropic described the campaigns as highly organised, utilising fake accounts, commercial proxy services, and evasion tactics to bypass its terms of service. Anthropic’s policy prohibits data harvesting for training and restricts access in China.
Safety and IP concerns raised over distillation
MiniMax reportedly ran the greatest effort, producing more than 13 million interactions, while Moonshot generated over 3.4 million and DeepSeek focused on extracting step-by-step reasoning traces. Anthropic stated that these operations targeted Claude’s strongest differentiators rather than general consumer queries, and in MiniMax’s case, the company detected activity in real time before the distilled model was released.
Anthropic warned that copied models could lack Claude’s built-in safety mechanisms, increasing risks of misuse such as generating harmful content, enabling cyberattacks, or facilitating other malicious applications. The company also tied the incident to broader geopolitical tensions, arguing that such practices undermine US export controls on advanced AI chips and hardware by enabling replication of frontier capabilities at lower cost.
“This was not casual or experimental,” Anthropic stated in its post. “These were industrial-scale campaigns” requiring significant resources, including access to advanced compute infrastructure.
Anthropic taking action to improve safety against distillation
Anthropic said it is bolstering detection systems, improving account verification, and collaborating with peers to combat similar threats. No immediate public responses from DeepSeek, Moonshot, or MiniMax were reported in connection with the allegations.
The accusations follow similar claims by OpenAI earlier in February 2026, which accused DeepSeek of targeting US models for distillation. The revelations come amid ongoing US debates over tightening AI chip export restrictions to China and highlight escalating friction in the global AI race.