Anthropic warns of 'industrial scale' AI data theft by Chinese firms

United States-based artificial intelligence firm Anthropic accused three Chinese unicorns— DeepSeek, Minimax and Moonshot AI — of having illegally extracted capabilities from its Claude model to advance their own systems, a new report has said. The report from CNN Business said that the US firm alleged that the theft through a process known as distillation raised national security concerns.

The modus-operandi of the alleged theft involved creation of around 24,000 fraudulent accounts to train Chinese models using over 16 million exchanges with Claude.

The company warned that models produced this way may lack the safety guardrails implemented by companies such as itself and thus could be used for cyberattacks, and biological weapons. These models could lead to “authoritarian governments to deploy frontier AI for offensive cyber operations, disinformation campaigns, and mass surveillance,” it said, warning that the “The window to act is narrow.”

CNN has reached out to DeepSeek, MiniMax and Moonshot AI for comment, the report said.

DeepSeek’s sudden rise to prominence in China, being dubbed as “AI tigers" had led to a sense of US export controls having failed.

The three unicorns currently rank among the top 15 models on the prominent Artificial Analysis leaderboard, the report added.