Researchers extract hidden AI reasoning and secrets via encrypted tokens

Original: Stealing Reasoning Traces from Proprietary LLM APIs

Source: stolen-thoughts.com ↗