Eigen RadarAI
Archive
Analysis

OpenAI discloses a campaign to extract protected reasoning

OpenAI publicly described a July campaign to extract protected internal reasoning from its models. It attributes a core cluster to people associated with Moonshot AI, the developer of Kimi, while leaving the identity of all operators uncertain. The company detailed account restrictions and technical defenses. Its request counts measure attempts, not confirmed successful extractions.

Artificial Intelligence··Midday
A security analyst beside computing hardware on a testing bench.

September disclosure describes July activity

OpenAI disclosed a coordinated attempt to extract protected internal reasoning from its models on September 30. It attributes a core cluster to individuals associated with Moonshot AI, the company behind the Kimi model. OpenAI says it remains unclear whether all operators belonged to one actor.[1], [2]

The earliest activity began on July 1, according to OpenAI. It spiked on July 24 and 25 with 16,000 requests from more than 4,000 users. The company says it disrupted a related cluster of more than 15,000 users by July 28. Those figures count attempted extractions, not necessarily successful ones.[1]

Conversation state became an extraction route

The described method copied encrypted reasoning from one conversation and asked a model in another conversation to decrypt and transcribe it. OpenAI says the operators did not break encryption, compromise a database or directly access stored user conversations. Independent researchers also reported related cross-model and conversation-compaction paths that the company confirmed were real.[1]

Account and output controls were tightened

OpenAI responded with account bans and restrictions, stronger signup controls and protections against replaying reasoning. It added checks on streamed output and shared findings with industry partners through the Frontier Model Forum. The September announcement describes the earlier activity and defenses; it does not report a new campaign beginning that month.[1]

References

  1. News sourceOpenAIOpenAI discloses reasoning-extraction campaign and alleges Moonshot links↩1↩2↩3↩4
  2. News sourceCellCogOpenAI reveals July campaign to extract hidden reasoning↩
AI safety and regulationChip wars and the supply chainThe frontier model race

More analysis from this desk

View all Brief analyses from the Artificial Intelligence desk →