On February 24, 2026, Anthropic, a U.S.-based AI company, lodged accusations against three Chinese AI labs—DeepSeek, Moonshot AI, and MiniMax—for allegedly creating over 24,000 fake accounts. These accounts were used to interact with Anthropic's Claude model more than 16 million times, with the labs illegally extracting its model capabilities through distillation techniques to enhance their own models. Anthropic asserted that these attacks were primarily aimed at exploiting Claude's unique strengths in agent-based reasoning, tool invocation, and code generation. The methods employed were highly similar across the labs, utilizing fake accounts and proxy services to avoid detection. Among the accused, MiniMax had the most extensive interaction, with over 13 million engagements. DeepSeek, on the other hand, generated chain-of-thought training data by prompting Claude to reveal its internal reasoning processes, including safety rewrites related to politically sensitive topics. Moonshot AI, with over 3.4 million interactions, concentrated on agent-based reasoning, tool usage, and computer vision capabilities.

Anthropic stated that it had confidently attributed these behaviors to the respective labs through technical evidence, such as IP address correlations and request metadata. The company called for a coordinated response from the industry, policymakers, and the global AI community to address such practices.

However, Anthropic's accusations quickly ignited controversy. Tesla CEO Elon Musk took to X (formerly Twitter) to retort, alleging that Anthropic itself had been involved in large-scale training data theft and had paid billions in settlements. Musk accused Anthropic of hypocrisy, using the phrase "the thief crying stop thief." Furthermore, netizens highlighted that Anthropic had trained Claude using copyrighted books and only paid royalties after facing legal action.

As of the time of reporting, DeepSeek, Moonshot AI, and MiniMax had not yet issued any responses regarding the matter.