Anthropic has revealed in a new report that several Chinese AI companies have been passing off answers from its Claude model as their own. According to the report, these developers were using Claude’s output and repackaging it under their own AI services.

What happened

The report calls out Moonshot, the company behind the Kimi assistant, for funneling real user queries into Claude and then serving up the answers as if they came from Kimi. In one documented episode, Anthropic tracked almost 300,000 queries run through this setup.

How the scheme worked

To get access to Claude, Moonshot allegedly set up a network of 5,380 fake accounts, mostly registered in Singapore and Japan. Anthropic says this infrastructure helped hide the true source of the traffic and scaled up the operation, letting them get tons of Claude’s answers without revealing where they were really coming from.

Other players and the scale

Anthropic also points to a similar playbook from DeepSeek. The report specifically highlights that Alibaba ran over 151,000,000 interactions with Claude, feeding those results into its own Qwen models. In total, Anthropic says seven Chinese AI companies were involved in these kinds of manipulations.

Anthropic describes this as a systematic effort to use Claude’s answers under other brands. The report names all the companies involved and breaks down the different ways they tapped into the Claude model.