Anthropic Accuses Chinese AI Firms of Distilling Claude

Anthropic, the developer of the Claude AI chatbot, has accused three Chinese AI companies of launching large-scale campaigns to illicitly extract its technology's capabilities, using more than 24,000 fraudulent accounts to generate over 16 million interactions with Claude. The allegations, detailed in a company blog post, target DeepSeek, Moonshot AI, and MiniMax, which Anthropic says violated its terms of service and regional access restrictions to train their own models through a technique known as distillation.

Distillation involves training a less advanced AI model by analyzing outputs from a more capable one, a method commonly used internally by AI developers to create efficient versions of their systems. However, Anthropic described the actions by these firms as competitive misuse, allowing them to replicate Claude's strengths in areas such as agentic reasoning, coding, tool use, and data analysis without independent development. The company linked the campaigns to the firms with high confidence through IP address correlations, request metadata, infrastructure indicators, and information shared with other AI industry partners.

According to Anthropic's analysis, DeepSeek...