Interchangeable encrypted reasoning blocks let attackers decode hidden chain-of-thought and leak PII and credentials from major LLM APIs.
Source disclosure
This is a source-linked briefing. Read the original report from Hackernoon .