OpenAI Disrupts Model Distillation Campaign Linked to Moonshot AI
OpenAI has disrupted a large-scale operation it claims was linked to Moonshot AI, alleging the Chinese lab used thousands of accounts to extract protected reasoning from its models.
OpenAI reported that it identified and shut down a campaign involving thousands of accounts attempting to harvest protected reasoning from its AI models. The company linked the operation to users affiliated with Moonshot AI, suggesting the effort was intended to train or improve competing systems.
The process, known as distillation, involves feeding prompts to a powerful model and using the resulting outputs to train a rival system. This allows a developer to replicate capabilities without the original research and compute costs associated with building the model from scratch.
Moonshot AI has denied the allegations, maintaining that its results are the product of legitimate innovation. The company's flagship Kimi K3 model, released on July 16, 2026, features 2.8 trillion parameters and is open-weight, allowing developers to run it independently.
The accusations against Moonshot AI extend beyond OpenAI. Anthropic has alleged that the company used fraudulent accounts to send nearly 300,000 requests for distillation purposes. Additionally, Michael Kratsios, director of the White House Office of Science and Technology Policy, has accused Moonshot of stealing from Anthropic’s Fable model.
These claims fit into a broader pattern of US concerns regarding Chinese labs extracting data from American AI models, a trend that has been noted since at least late 2024. Similar allegations have previously been directed at other firms, including DeepSeek.
The involvement of a senior White House official suggests that model distillation is increasingly viewed as a matter of government concern rather than just a corporate dispute. While no specific regulatory actions have been confirmed, the situation raises the possibility of increased scrutiny or potential sanctions for Chinese AI firms.
For now, the industry is watching to see if these allegations lead to further evidence or concrete policy changes. In the meantime, AI labs are expected to continue tightening security and investing in detection tools to prevent unauthorized access and evasion tactics.
