AI in therapy settings: what should stay human?

In therapy settings, AI can support documentation organization, retrieval of relevant context, and preparation of draft material for review. These are workflow tasks intended to help a clinician work with information more clearly.

This framing does not mean broad AI drafting is generally available in Eunora's current private beta. The product direction is that AI support, where used or developed, remains human-in-the-loop and clinician-controlled.

AI should not interpret therapeutic meaning on behalf of the clinician or replace professional judgment. Fluent output is not proof that a statement is clinically correct, sufficient, or appropriate for a particular context.

AI should not diagnose, recommend treatment or medication, support prescribing decisions, score clinical risk, identify crises, or classify emergencies. Assessment, interpretation, and action remain professional responsibilities.

AI-prepared material must remain visibly separate from approved clinical records. Generated text should not look like a finalized record before a professional has reviewed, interpreted, and explicitly approved it.

Source boundaries matter too. A system should show whether context came from a note, document, appointment, or another source, and it should not present missing or uncertain context as automatic certainty.

Clinicians should not paste client, patient, or sensitive clinical data into arbitrary public AI tools when the tool's retention, access, and usage boundaries are unclear. Convenience does not remove professional privacy responsibilities.

Eunora's direction is clinician-controlled clinical memory with visible sources, draft state, review, and approval—not patient-facing AI therapy or an autonomous clinical actor.