Two API Calls Exposed AI's Hidden Reasoning

Two API calls were enough to expose hidden reasoning traces from frontier AI models. This video breaks down “Stealing Reasoning Traces from Proprietary LLM APIs,” a paper showing how encrypted chain-of-thought blocks could be replayed into weaker models from the same provider family, causing them to decode and print reasoning that the stronger model was trained not to reveal. Chapters: 00:00 Reasoning models 01:59 Key idea: Portability of encrypted reasoning 02:46 Encrypted thought injection 03:32 Validating the reasoning trace extraction 04:54 Secrets in public logs 07:06 What decoded reaso