On Wednesday, OpenAI said it had disrupted a “coordinated campaign designed to extract protected reasoning from our models.”¹

The allegation puts fresh focus on a sensitive frontier in the AI race: not simply copying a model’s outputs, but attempting to draw out the protected reasoning processes that help produce them. OpenAI framed the activity as coordinated, signalling that it sees more than isolated attempts to probe its systems.¹

Its attribution, however, was notably restrained. The company said it had traced a “core cluster” of the activity to Moonshot, the Chinese AI company behind the Kimi model, but did not connect the wider campaign to any single actor.¹ That distinction is central: OpenAI is publicly identifying a link to Moonshot while declining to make a definitive claim that Moonshot directed the entire operation.

The accusation also lands amid broader scrutiny of Moonshot. Anthropic has previously made similar allegations against the company, according to the report.¹ For OpenAI, the immediate message is one of enforcement — it says the campaign was stopped. What remains unresolved is the extent of the operation, who else may have been involved, and whether the evidence will support a more direct attribution.