China’s Moonshot AI recently released its latest large language model, Kimi K3, claiming performance comparable to leading models such as Claude and GPT-5-class systems. Chinese media have promoted Kimi as a domestic alternative to Claude and ChatGPT. However, user testing has reignited questions over whether the model was trained using outputs generated by Anthropic’s Claude.
Screenshots circulating online show a user said, “hey kimi hows it going!” and Kimi K3 responded with: “Hey! Doing well, thanks for asking! Quick heads up though—I’m Claude, not Kimi. Kimi is a different AI assistant (made by Moonshot AI). Easy mix-up, no worries!”
Similar behavior was previously observed in Kimi K2.5, which occasionally identified itself as Claude after extended conversations or when asked directly about its identity. The behavior resembles patterns described by Anthropic, that models distilled or trained on Claude-generated outputs may inherit Claude’s self-identification and response characteristics.
The incident has renewed scrutiny of Chinese AI developers’ use of model distillation—a technique that trains smaller models on the outputs of more advanced systems to reduce development costs and accelerate deployment. OpenAI has previously accused DeepSeek of improperly using ChatGPT-generated data. In addition, Anthropic alleged in a June letter to the U.S. Senate Banking Committee that operators linked to Alibaba’s Qwen AI laboratory used nearly 25,000 fraudulent accounts to generate approximately 28.8 million interactions with Claude in an effort to harvest model outputs. The repeated identity confusion exhibited by Kimi has fueled further speculation that Chinese AI developers may be relying extensively on distilled outputs from leading U.S. models.
Source: NTDTV, July 19, 2026
https://www.ntdtv.com/b5/2026/07/19/a104116746.html