Anthropic has accused Chinese artificial intelligence companies DeepSeek, Moonshot AI, and MiniMax of using tens of thousands of fraudulent accounts to extract capabilities from its Claude model to enhance their own systems. The company said the activity generated more than 16 million interactions targeting advanced features such as agentic reasoning, tool use, and coding.
Anthropic reported that the efforts varied by organization, with large volumes of exchanges aimed at improving logic, alignment, computer-use agents, and orchestration capabilities. The company warned that large-scale distillation could undermine safeguards built into frontier AI models and amplify national security risks if replicated without controls.
Anthropic stated it is strengthening defenses and called for coordinated action among technology firms, cloud providers, and policymakers as global competition in AI intensifies.
