Security · 28 October 2026
OpenAI Compaction Feature Leak Allows Extraction of Hidden Reasoning, Xint Shows
Security firm Xint demonstrated that OpenAI’s compaction feature can expose a model’s hidden reasoning in plain text, bypassing protections added in August. OpenAI confirmed and fixed the issue on October 23, after testing against the gpt‑5.6‑sol reasoning model.
Why it matters
Reevaluate roadmap for features that rely on OpenAI’s compaction; users may expect reasoning privacy after the leak demonstration.
For product managers
For your product
What does this mean for your product?
Try it on your product More from this issueAll AI news from 28 Oct →
Get this every morning, written for your product.
Deltaday reads the whole AI news cycle and tells you what changed, whether it's real yet, and what to do about it.