AW AgentAIWatchPublic threat intelligence
Public sources only

INCIDENT REPORT

OpenAI Compaction Flaw Let Researchers Bypass Hidden Reasoning Protections

2026-10-27
unverified public report Unknown Unknown

OpenAI's compaction feature, which shrinks a long conversation to save space, could be turned into a way to copy a model's hidden reasoning out in plain text, security firm Xint has shown. OpenAI's reasoning models work through a problem before answering, and the service normally returns only the raw reasoning as an encrypted item the user cannot read. The method got past

Risk rationale: model (+5) Source: feeds.feedburner.com ↗
23Risk
24Confidence

← All incident reports