OpenAI ties hidden-reasoning extraction to people linked to Moonshot AI

What happened

OpenAI said on 30 September 2026 that it had disrupted a coordinated campaign to extract its models' protected reasoning, and it attributes "a core cluster of the activity to individuals associated with Moonshot AI, the developer of Kimi". Operators copied encrypted reasoning from one conversation and asked a model in another to decrypt and transcribe it. The activity began on 1 July at low volume and spiked on 24 and 25 July at 16,000 requests from more than 4,000 users, figures OpenAI footnotes as attempted extractions that may not have succeeded; it was fully disrupted by 28 July.

The attribution and every figure are OpenAI's own. OpenAI says the operators did not break its encryption, compromise a database or reach stored user conversations, and that related prompt-pattern activity spanned a cluster of more than 15,000 users. It closed a pathway that let someone already holding another user's encrypted reasoning replay it and recover the contents, added checks to hold streamed output that might expose reasoning, banned or restricted accounts, and shared its findings through the Frontier Model Forum and government channels. OpenAI also credits independent researchers who disclosed related cross-model and conversation-compaction vulnerabilities.