kapynResearch

"But marinade" and leaked passwords are what researchers found in ChatGPT's hidden reasoning

Researchers extract encrypted model reasoning traces and discover exposed credentials in public sessions across major AI APIs. The security flaw reveals that user-facing reasoning summaries often obscure the underlying processing steps of frontier models. This discovery highlights critical privacy and data leakage risks in hosted LLM platforms.

The Decoder·Aug 11, 2026

Opening Kapyn…