Covered by 2 outlets — first reported by The Hacker News, Oct 27, 08:22 UTC
OpenAI Compaction Flaw Let Researchers Bypass Hidden Reasoning Protections
MentionsOpenAI
Summary
Security firm Xint showed that a feature in OpenAI's service that shrinks long conversations, called compaction, could be used to copy out a model's hidden reasoning in plain text. The Hacker News reports the service normally returns that reasoning only as an encrypted item the user cannot read. Infosecurity Magazine reports the flaw in OpenAI's API led to a fix and a $6000 bug bounty reward.
AI summary by the SecurityTabs newsroom, checked against The Hacker News, Infosecurity Magazine. Read the outlets below for the full story.
Continue reading
2 outlets, in the order they reported it.
The Hacker NewsFirst report
OpenAI Compaction Flaw Let Researchers Bypass Hidden Reasoning Protections
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.
Infosecurity Magazine
OpenAI Compaction Flaw Exposed AI Reasoning to Researchers
Researchers uncovered an OpenAI API flaw that exposed hidden AI reasoning traces, prompting a fix and a $6000 bug bounty reward