OpenAI accuses Moonshot of coordinated campaign to extract model reasoning

OpenAI announced on September 30, 2026, that it disrupted a coordinated campaign in July 2026 aimed at extracting protected reasoning from its artificial intelligence models. While OpenAI could not link the entire campaign to a single actor, it traced a core cluster of the activity to individuals associated with Chinese rival Moonshot AI, developer of the Kimi model. The campaign involved 'adversarial distillation' to copy advanced capabilities, raising significant safety and national security concerns.

Attribution of the extraction campaign

• ▪Anthropic previously accused several Chinese artificial intelligence developers, including Moonshot AI and Alibaba, of secretly using its Claude model to help train their own systems
• ▪OpenAI stated that it was unclear whether all the operators involved in the extraction campaign attributed to Moonshot AI were linked to a single actor

Risks of model reasoning extraction

• ▪OpenAI stated that extracting its models' reasoning could allow competitors to reproduce advanced capabilities without making the same investment in developing and safeguarding frontier models
• ▪OpenAI warned that the unauthorized extraction of protected reasoning from its frontier artificial intelligence models poses potential safety and national security risks

Debatable claims

• ▪Coordinated model extraction campaigns constitute a form of cyberattack
• ▪Adversarial distillation of frontier AI models poses a genuine threat to national security
• ▪OpenAI's public attribution of the extraction campaign to Moonshot AI is justified
• ▪Using a competitor's AI model outputs to train a rival system is ethically acceptable