On February 23, 2026, U.S. AI company Anthropic issued a report alleging that Chinese AI companies DeepSeek, Moonshot AI, and MiniMax had engaged in over 16 million interactions with its model Claude via roughly 24,000 fake accounts. The purpose, Anthropic claimed, was to systematically extract Claude’s core functionalities—including reasoning, tool invocation, and programming—to train and enhance their own models. According to the report, these interactions were designed to disperse traffic and evade detection through a “Hydra architecture,” amounting to what Anthropic described as an “industrial-scale distillation attack.” Specifically, MiniMax was said to have conducted over 13 million interactions, primarily targeting tool orchestration and agent coding; Moonshot AI allegedly carried out over 3.4 million interactions, focusing on agent reasoning, programming, and related areas; while DeepSeek reportedly had over 150,000 interactions, with an emphasis on acquiring “chain-of-thought” data for reinforcement learning training. Anthropic stressed that while distillation techniques are not inherently illegal, the unauthorized extraction of a competitor’s capabilities poses serious risks. These include potentially undermining safeguards against bioweapons development, facilitating malicious cyberattacks, and even escalating to national security threats.

However, Anthropic’s accusations have been met with widespread skepticism. Elon Musk ridiculed Anthropic, suggesting it was “stealing data from human programmers themselves” and pointed out that Anthropic had previously paid billions in compensation for large-scale theft of training data. Industry experts also weighed in, arguing that Anthropic’s evidence was insufficient, its attribution methods crude, and that it was applying a double standard. They noted that global AI giants routinely train models using copyrighted works without permission, yet accuse Chinese companies of “illegal distillation.” Furthermore, Anthropic faced criticism for politicizing technical disputes, attempting to pressure the U.S. government into strengthening AI chip export controls against China and stifling Chinese AI development by exaggerating national security risks. As of now, DeepSeek, Moonshot AI, and MiniMax have not issued any responses to these allegations.