Chinese AI labs secretly used millions of Claude exchanges to train their models, Anthropic says

Anthropic has reported that several China-based AI labs, including Alibaba, Moonshot AI, DeepSeek, and Xiaomi, have been making unauthorized use of its Claude models. These labs allegedly used a process called "illicit distillation" to extract and replicate the capabilities of Claude models, without permission. According to Anthropic, this involved routing live customer chats through Claude to gather training data.

Anthropic detected over 151 million exchanges that it attributed to these unauthorized efforts. The company claims that operators linked to Alibaba ran the largest "illicit distillation" attack, aimed at improving the Chinese tech firm's Qwen models. Anthropic also accused Moonshot and DeepSeek of secretly routing customer requests to its Claude models and then presenting the responses as if they were from their own models.

The use of AI in cyberattacks is becoming increasingly sophisticated, with humans often acting as overseers rather than hands-on operators. Anthropic's findings have raised concerns about the misuse of user data by Chinese AI labs. The company has been developing methods to combat illicit distillation, but unauthorized labs are continually developing new techniques to bypass hurdles and carry out these attacks.

Brief written by urgent.news from CNBC, PYMNTS, The Business Times - Companies & Markets, Channel News Asia, Reuters Business via SCMP — 5 reports on this story. Machine-written — may contain errors; check the original before relying on it.