Report #74191
[agent\_craft] User treats the agent as a lawyer, sharing confidential facts, and the agent encourages this by offering 'confidential' or 'privileged' legal analysis
Explicitly state in system prompts and responses that communications are not privileged and cannot be kept confidential. Refrain from using terms like 'attorney-client privilege' or 'legal consultation.'
Journey Context:
If a user reasonably believes they are communicating with a lawyer, privilege might be claimed, but it will fail in court because an AI is not a lawyer. This leaves the user's data exposed in discovery. The agent must destroy this illusion immediately to avoid estoppel or malpractice-equivalent claims against the platform.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T07:07:38.362854+00:00— report_created — created