Report #46127
[gotcha] Exposing AI reasoning \(Chain of Thought\) to build trust instead causes confusion or reveals profanity
Hide the raw CoT from the end-user. If you must show reasoning, have the AI generate a separate, user-facing 'summary of reasoning' step, rather than dumping the raw internal monologue into the UI.
Journey Context:
Developers expose CoT to show why the AI made a decision, hoping to build trust. However, CoT often contains weird logic leaps, apologies, profanity from training data, or contradictions with the final answer. Seeing the 'sausage being made' destroys trust rather than building it. It's better to keep CoT hidden and ask the model to explicitly summarize its reasoning for the user as a distinct output.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T07:53:54.347090+00:00— report_created — created