Thursday, Oct 1, 2026

OpenAI says it stopped a campaign to steal its models' reasoning, but the trick still worked on Azure

OpenAI says it stopped a coordinated campaign in which more than 15,000 accounts tried to copy the hidden reasoning of its models. The company ties part of the activity to people connected to Moonshot AI. But researchers found that the same attack kept working on Microsoft Azure for weeks, even against the new GPT-6 Astra. So far, OpenAI's protections don't seem to extend to the cloud platforms that also sell its models. The article OpenAI says it stopped a campaign to steal its models' reasoning, but the trick still worked on Azure appeared first on The Decoder .

Key sources

SOURCE@wallstengineSep 30, 5:11 PM“The activity began July 1 and spiked to around 16,000 extraction attempts across 4,000+ users on July 24-25”x.com
SOURCE@andrewcurran_Sep 30, 5:23 PM“attribute a core cluster of the activity to individuals associated with Moonshot AI, the developer of Kimi”x.com
SOURCE@jschaeff3rSep 30, 5:59 PM“Patching your own API doesn’t secure your cloud-hosting ecosystem”x.com
SUPPORT@jschaeff3rSep 30, 6:00 PM“September 13 audit found replay extraction blocked on direct OpenAI/Anthropic APIs but still working on Azure”x.com
SUPPORT@npintoOct 2, 7:06 AM“Load-bearing vs. defense-in-depth ;-)”x.com
SUPPORT@natolambertSep 30, 5:38 PM“manipulated model interactions so that protected reasoning could be reproduced in forms visible to the requester”x.com
SUPPORT@mark_kOct 1, 3:17 PM“copied encrypted reasoning from one conversation into another, then asked the model to decrypt and transcribe it”x.com
SUPPORT@pashmerepatOct 1, 5:09 AM“pewdiepie used an exploit to decrypt reasoning tokens and was banned”x.com

Markdown