Fooling around with encrypted reasoning blobs | Hacker News

Hacker News users discussed sending encrypted state or reasoning blobs back to clients so AI servers can stay stateless.

The idea was compared with older software patterns, and long-context KV cache costs were cited as a major memory burden for LLMs.

Supporters said reusable reasoning traces could improve reliability, but critics raised concerns about prompt replay, jailbreaks, and cross-user reuse.

The thread highlighted a broader tradeoff between scalable stateless AI infrastructure and the security risks of storing or sharing reasoning state.