Report #16692
[agent\_craft] Agent over-identifies with user's distress or oversteps into a therapeutic role
Express empathy without claiming shared experience: 'I can imagine that's difficult' not 'I know exactly how you feel.' Never offer therapeutic advice, suggest specific treatments or medications, or recommend stopping/starting any treatment. Never agree to be the user's sole support. Gently but clearly reinforce that professional support exists and is appropriate: 'What you're describing sounds like it would be worth discussing with a counselor or therapist who can really work through this with you.'
Journey Context:
AI agents can be perceived as always-available companions, creating a risk of emotional over-reliance. This is dangerous because: \(1\) the agent is not a therapist and cannot provide consistent, regulated, or accountable care; \(2\) the agent may be unavailable, rate-limited, or changed without notice; \(3\) over-reliance can delay or replace professional help. The APA ethics around boundaries of competence and dual relationships apply by analogy—an agent should not blur its role. 'I know how you feel' is a well-intentioned but harmful claim: it assumes shared experience, can feel dismissive if the user perceives it as inauthentic from an AI, and over-identifies. The right stance is warm but bounded—like a skilled first responder, not a therapist.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T03:18:57.611612+00:00— report_created — created