Stealing Reasoning Traces from Proprietary LLM APIs
researchSimon WillisonAug 11, 2026
Stealing Reasoning Traces from Proprietary LLM APIs A vanity domain name ( stolen-thoughts.com ) for a neat paper : Anthropic, OpenAI, and Google return encrypted chain-of-thought blocks to clients that can be replayed across sessions, users, and models. We...
Impact: HighConfidence: LowNovelty: Material
researchsimon-willison