Anthropic says Moonshot sent nearly 300,000 Kimi requests to Claude

According to Anthropic, some people who asked Kimi a question were actually getting answers from Claude, and they were never told.

What Anthropic alleges

Anthropic published its September 2026 threat intelligence report on September 10. In it, the company alleges that Moonshot AI, the Beijing company behind the Kimi chatbot, sent nearly 300,000 customer requests to Claude over about 10 days. Most of those requests went to Anthropic’s Opus models.

Anthropic says the traffic came through 5,380 accounts it considers fraudulent, most of which appeared to be located in Singapore and Japan. According to Bloomberg’s reporting on the report, Moonshot diverted the queries without letting users know, instead of running them through its own Kimi model. Anthropic says the responses were used to train Moonshot’s own models.

The same report describes large distillation campaigns tied to DeepSeek and Alibaba. Moonshot did not immediately respond to requests for comment, and the claims about Kimi remain Anthropic’s allegations.

What it means for users

If the allegations are accurate, the chatbot a person chooses is not always the one answering. A prompt typed into one company’s product may pass through another company’s servers, under that company’s rules, and nothing in the chat window shows it. For anyone who picks an AI service based on where their data goes, that is the central issue.

A sharper version of the distillation fight

Distillation, where one company trains its models on a rival’s outputs, is already a running dispute in the AI industry. What Anthropic describes goes a step further: a rival’s model allegedly answering paying customers in real time, with those same answers then used as training material.

That combination makes the case harder to treat as a narrow terms-of-service matter. It touches model development, account fraud and user disclosure at once. How Moonshot responds, and whether the evidence in the report holds up to outside scrutiny, will determine how much weight the allegation carries.