OpenAI Compaction Flaw Let Researchers Bypass Hidden Reasoning Protections

Reported by The Hacker News ([email protected] (The Hacker News))

Summary by Cyber Security News27 Oct 2026, 09:22 CET

About this summary. This is a short excerpt from the description that The Hacker News published with its article. Cyber Security News did not report or verify the underlying story. Read the original: https://thehackernews.com/2026/10/openai-compaction-flaw-let-researchers.html

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

Read the full article at The Hacker News →

Source attribution: headline and facts are from The Hacker News (thehackernews.com). Summary method: excerpt of the source description. See our source attribution policy.