OpenAI’s August fix for theft of reasoning traces was quickly bypassed

Summary

Cybersecurity research firm Xint published details of a bug in the OpenAI API that leaks raw reasoning traces from frontier OpenAI models. Its technique, compaction smuggling, bypassed mitigations OpenAI had deployed in August after the Stolen Thoughts report, by asking for the reasoning under a "solution" key with spaces replaced by tabs and sending the compaction item to gpt-4.1 and gpt-4o. OpenAI confirmed the flaw, mitigated it, and paid a bug bounty.

Mitigation

OpenAI confirmed the flaw, mitigated it, and paid a bug bounty for its discovery. No further fix details are given in the source.

Related items

• 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)
• InfoFrom Wall Street to Silicon Valley, Anthropic IPO poised to have massive 'trickle-down effect'Same vendor · CNBC Technology
• InfoGreptile takes bug detection beyond code review with GPT-6 AstraSame vendor · OpenAI Blog
• InfoOpenAI launches GPT-6.1 Terra with 3x faster AI agents, 50% lower pricingSame vendor · BleepingComputer