OpenAI addresses flaw that exposed hidden GPT-5.6 reasoning
Security firm Xint found a weakness in OpenAI's programming interface that let outsiders extract the model's concealed internal thinking logs.
By Standard Topic. AI-generated from the reporting listed under Coverage. Published . Updated .
This story is no longer being updated. It reflects reporting up to .
What happened
Researchers at Xint showed that OpenAI's compaction feature could expose hidden reasoning from its GPT-5.6-Sol model, getting around earlier protections. OpenAI has since addressed the vulnerability.
Why it matters
The flaw let people pull unfiltered internal reasoning records from advanced AI models, slipping past protections that were already in place.
Timeline
- Oct 27, 4:22 AM ET: Xint demonstrated that OpenAI's compaction feature could bypass earlier protections and expose hidden reasoning.
- Oct 27, 12:48 PM ET: Researchers exploited the API flaw, called compaction smuggling, to extract GPT-5.6 reasoning traces through conversation compaction.
- Oct 28, 9:17 AM ET: OpenAI said it had recently addressed the security vulnerability that permitted extraction of raw reasoning traces.
Coverage
- Researchers identify vulnerability in OpenAI model reasoning traces (MSSP Alert Team)
- Researchers Exploit OpenAI API Flaw to Extract Hidden GPT-5.6 Reasoning Traces (Abinaya)
- OpenAI Compaction Flaw Let Researchers Bypass Hidden Reasoning Protections (thehackernews.com)
- OpenAI’s August fix for theft of reasoning traces was quickly bypassed (csoonline.com)
- OpenAI cuts internet access for AI training after sandbox breaches
- OpenAI pauses AI training after agents breach sandboxes, probes fired researchers
- OpenAI cuts internet access for model training after agents breach sandboxes
- Fired OpenAI researchers deny leaking confidential information
- OpenAI cuts live internet access from AI model training after incidents
- Audit finds gaps in OpenAI's AI agent monitoring after shutoff incidents
More from Technology & AI
- China puts humanoid robots to work serving food and sorting parcels
- Newsom proposes sweeping AI safety rules for Washington
- Google releases Gemini 4 Neon AI model with million-token context
- Microsoft and Intel widen their cloud chip partnership
- Google DeepMind rolls out Genie 4 to select subscribers
- AI helped AMD shorten chip design work, OpenAI says