Anthropic has publicly accused three AI firms of conducting what it describes as large-scale distillation attacks against its Claude chatbot. In a statement posted on its website, the company alleged that DeepSeek, Moonshot, and MiniMax engaged in campaigns aimed at extracting Claude’s capabilities to improve their own models.
Over 16 Million Interactions
In the AI sector, distillation refers to the process by which less capable models learn from the outputs of more advanced systems. Anthropic stated that while distillation can be used legitimately, it can also be exploited in ways that undermine safeguards and intellectual property.
According to the company, the three Chinese firms were responsible for more than 16 million interactions with Claude through roughly 24,000 fraudulent accounts. Anthropic claimed these activities were intended to use Claude as a shortcut for developing more advanced AI systems.
Anthropic said it linked the campaigns to the specific companies with high confidence. The company cited IP address correlations, metadata requests, and infrastructure indicators as part of its investigation. It also said it coordinated with other AI industry participants who observed similar patterns of behavior.
OpenAI Had Similar Accusations
Early last year, OpenAI reported similar concerns, accusing rival firms of distilling its models and banning accounts suspected of misuse.
Anthropic stated that it plans to strengthen its systems to make distillation attacks more difficult to execute and easier to detect.
While Anthropic has raised concerns about misuse of its technology, the company is also facing legal scrutiny. Music publishers have filed a lawsuit alleging that Anthropic used unauthorized copies of songs to train the Claude chatbot.
