OpenAI Compaction Flaw Let Researchers Bypass Hidden Reasoning Protections

Summary

Security firm Xint showed that OpenAI's compaction feature, which shrinks long conversations, could be used to copy a model's hidden reasoning out as plain text. The method bypassed the cross-model replay block OpenAI added in August, because compaction runs on the same model that produced the reasoning. OpenAI confirmed and fixed the problem on October 23, and Xint tested only gpt-5.6-sol.

Mitigation

OpenAI fixed the issue on its side on October 23; developers who build on the API do not need to change anything.

Related items

• InfoAnthropic's IPO is risky businessSame vendor · CNBC Technology
• InfoData center backlash is uniting AmericaSame vendor · The Verge (AI)
• MediumOpenAI’s August fix for theft of reasoning traces was quickly bypassedSame vendor · CSO Online
• InfoGPT-6.1 TerraSame vendor · Simon Willison's Weblog
• InfoOpenAI announces three big science discoveries and a new way to work with academicsSame vendor · The Verge (AI)