Encrypted AI reasoning blocks may be decodable across sessions and models

Stolen Thoughts says encrypted reasoning blocks returned by OpenAI, Anthropic and Google APIs are interchangeable across sessions, users and models, allowing hidden reasoning to be decoded. It says it reconstructed 315,320 blocks from 6,708 publicly available agent trajectories and recovered 704 distinct privacy artifacts from non-benchmark user sessions; 64 appeared only in reasoning blocks. In an October 18 reply, a commenter pointed to a September 30 update and said the issue “keeps happening.”