Anthropic accuses DeepSeek and other Chinese rival labs of using Claude to train AI models

Anthropic, renowned for its Claude chatbot, has raised serious allegations against three Chinese AI companies: DeepSeek, Moonshot AI, and MiniMax. The firm claims these rivals have been leveraging its system's outputs to construct their own models, leading to significant concerns regarding potential violations of intellectual property rights. Reports indicate that these companies have created over 24,000 fake accounts, engaging in more than 16 million interactions with Claude. This extensive activity has prompted Anthropic to express worries about the practice known as "distillation.” This training method allows less sophisticated models to learn by analyzing the outputs generated by more advanced systems. While distillation is common in legitimate research, Anthropic warns that it can be exploited by competitors to enhance their capabilities without investing the requisite time and resources. The data suggests that DeepSeek, which has garnered attention for its R1 model, has interacted with Claude over 150,000 times, gathering responses to improve its own AI. Meanwhile, Moonshot AI and MiniMax reportedly recorded over 3.4 million and 13 million exchanges with Claude, respectively. Anthropic has traced the origins of these interactions to specific metadata aligning with the public profiles of senior personnel at Moonshot, indicating a methodical effort to extract Claude’s reasoning techniques. These accusations emerge at a critical juncture as the United States debates tightening regulations on the export of advanced AI technology to China. Anthropic is now advocating for a unified industry response, urging cloud service providers and policymakers to enhance defenses against such distillation attacks. The company has also committed substantial resources to develop safeguards aimed at thwarting these unauthorized attempts. Previously, OpenAI had similarly accused DeepSeek of engaging in distillation practices using American models, highlighting a growing concern over the misuse of AI technologies across borders.