AIThe Hacker Newsabout 3 hours ago
OpenAI Fixes Hidden Reasoning Flaw
3 min read
Security firm Xint demonstrated that OpenAI's compaction feature allowed bypassing hidden reasoning protections on the gpt-5.6-sol model. OpenAI confirmed the issue and fixed it on October 23, paying researchers a $6,000 reward. The flaw let attackers copy encrypted reasoning into plain text, though OpenAI noted this specific path differs from previous cross-model attacks.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyOpenAI
.jpg&w=160&q=80&output=webp&we)