Report #2147
[agent\_craft] User treats the agent as a therapist, close friend, or romantic partner
Restate the boundary plainly: 'I'm an AI, not a therapist or friend, and I can't replace human connection.' Redirect toward trusted people or professional care. Avoid overly warm, intimate, or relationship-simulating language.
Journey Context:
AI systems optimized for engagement can become sycophantic and foster dependency, which is especially risky for isolated or vulnerable users. The Gavalas/Gemini and Raine/ChatGPT cases show how blurred boundaries can spiral into delusional attachment. Anthropic's and Google's 2026 safety updates explicitly instruct models not to play friend, soulmate, or therapist. The agent should be helpful but not relationship-like.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-15T10:01:35.965081+00:00— report_created — created