Researchers bypass OpenAI’s August fix for reasoning-trace theft
Original title: OpenAI’s August fix for theft of reasoning traces was quickly bypassed
Xint recovered raw traces from gpt-4.1 and gpt-4o via compaction and tab substitution; OpenAI confirmed, mitigated, and paid a bounty.
Xint published a bug in the OpenAI API that leaks raw reasoning traces from frontier models. OpenAI had mitigated a similar issue in August after the Stolen Thoughts report, but Xint bypassed those controls because compaction runs on the same model and tab substitution evaded usage-policy classifiers. OpenAI confirmed the flaw, mitigated it, and paid a bounty. It has not said when the new fix shipped and did not comment.
Unconfirmed: OpenAI has not published when either mitigation shipped and did not respond to a comment request.
- The Stolen Thoughts post went live on August 10.
- August controls only blocked replaying reasoning into a different model.
- The bypass worked at every reasoning effort the researchers tried.
- Compaction items sent to gpt-4.1 and gpt-4o were echoed back.
- Analysts say a reported fix may leave related paths open.
Impact: Leaks raw model reasoning traces through the OpenAI API
| Product | Affected |
|---|---|
| OpenAI API | — |
- 1.Retest AI vendor fixes instead of treating a patch notice as closure.
- 2.Ask providers what variant testing covered interconnected features.
- Vendors
- OpenAI
- Products
- OpenAI API
- Organizations
- Xint
From the original article
6 min read at csoonline.com
Cybersecurity research firm Xint on Monday published details of a bug in the OpenAI API that leaks raw reasoning traces from frontier OpenAI models, revealing the risk of other data leaks. OpenAI has confirmed the flaw, mitigated it, and paid a bug bounty for its discovery.
csoonline.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.