AI · 5h ago
On October 23, 2026, OpenAI confirmed Xint had bypassed its earlier mitigations to transfer reasoning traces between models through context compaction, fixed the issue, and awarded Xint a $6,000 bounty.
As a conversation grows, compaction compresses its history into a summary to fit the model’s context limit. Xint found it could shape that summary to carry one model’s reasoning, then pass it to another model and ask it to reproduce the contents. That creates a route for reasoning to cross model boundaries; it is not a theft of model weights.
For teams using OpenAI APIs, context summaries belong inside the trust boundary: they can carry proprietary behavior between models. OpenAI says it fixed this path, but Xint’s report does not show that anyone used the traces to train a rival model or establish whether other providers have the same weakness.
2 sources covering this story
OpenAI Compaction Flaw Let Researchers Bypass Hidden Reasoning Protections
Xint demonstrated how OpenAI's compaction feature could expose GPT-5.6-Sol's hidden reasoning, bypassing earlier protections.
Stolen AI Reasoning Could Help China Build Cheaper Rivals
Xint says a flaw in OpenAI’s compaction process let researchers move reasoning traces between models, potentially making it cheaper to distill a rival AI.
Part of the PlainSec briefing for 2026-10-27