When you use an AI app, you assume you are talking to the AI it advertises. A new report from Anthropic suggests that assumption may not always hold, and the accusation at the center of it could reshape how much you trust any AI product you did not build yourself.
On September 10, 2026, Anthropic published a report titled Detecting and countering misuse of AI, and one chapter made waves across the industry. In it, Anthropic accuses Moonshot AI, the Chinese company behind the popular Kimi app, of secretly routing its users’ requests to Claude and passing the answers off as Kimi’s own.
What Anthropic says happened
According to the report, Moonshot silently forwarded customer requests that were meant for its own Kimi models to Anthropic’s Claude instead, displayed Claude’s responses to users as if Kimi had generated them, and saved those exchanges to help train its own models. That last step is a technique known as distillation, where one model learns from the outputs of a stronger one.
The scale is significant. In one instance, Anthropic said it detected nearly 300,000 customer requests that Moonshot routed primarily to Anthropic’s Opus model. Because Anthropic does not allow its technology to be accessed from inside China, Moonshot allegedly used 5,380 fraudulent accounts to get around the block, most of them appearing to be based in Singapore and Japan. Anthropic’s report also pointed to DeepSeek, another Chinese AI company, and broader reporting has tied large volumes of activity to other players in the region. Moonshot declined to comment.
Why the cheaper angle stings
Kimi has built its appeal partly on price. It is positioned as a low-cost alternative to premium Western models like Claude. If Anthropic’s allegation is accurate, the irony is sharp: users chose the cheaper option and may have been served the very model they were trying to save money on, without knowing it and without Anthropic’s permission.
That is the part that should give any buyer pause. It is not just a dispute between two companies. It is a reminder that when you use a third-party AI product, you often cannot verify which model is actually answering you. The entire industry runs on a kind of black-box trust, and this case pokes a hole in it.
What it means for your business
If your company relies on an AI vendor, this is a prompt to ask harder questions. Which model actually powers the product? Where is your data being sent and stored? Could your inputs be used to train someone else’s model? These were once niche concerns for compliance teams. After this week, they are basic due diligence.
The Kimi case also lands at a sensitive moment for the global AI race. Accusations that Chinese firms are quietly leaning on American models feed directly into the larger debate about competition, security, and who really leads in AI. Expect this to become a talking point well beyond the technical community.
For now, the lesson is practical and a little unsettling. The AI you are paying for might not be the AI you are getting. In a market built on trust in the black box, that is a problem the whole industry will have to answer for.
