Skip to content

Topic

Chain-of-thought visibility

Access to a model's intermediate reasoning traces, used by developers and reviewers to detect deception, misuse or misalignment that finished outputs would hide.

Current clusters