ModelsOct 1, 2026 · 19:058 sources

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

From the-decoder.com

Image: the-decoder.com
AI;DR

OpenAI reported it stopped a coordinated campaign involving more than 15,000 accounts attempting to copy the hidden reasoning of its models. While OpenAI tied part of the activity to people connected to Moonshot AI, researchers found the attack continued to work on Microsoft Azure for weeks, including against GPT-6 Astra.

Read the sourceSign in to suggest

Sources

+ 5 more
  1. SUPPORT@jschaeff3r“September 13 audit found replay extraction blocked on direct OpenAI/Anthropic APIs but still working on Azure”x.com
  2. SUPPORT@npinto“Load-bearing vs. defense-in-depth ;-)”x.com
  3. SUPPORT@natolambert“manipulated model interactions so that protected reasoning could be reproduced in forms visible to the requester”x.com
  4. SUPPORT@mark_k“copied encrypted reasoning from one conversation into another, then asked the model to decrypt and transcribe it”x.com
  5. SUPPORT@pashmerepat“pewdiepie used an exploit to decrypt reasoning tokens and was banned”x.com

One ranked edition, every morning

The stories that matter today. Free, cancel anytime.

Telegram @aidr_today