AI delivers or is offered as therapy to the public without a licensed professional conducting it

Have a licensed clinician conduct every therapy engagement with AI output only as a reviewed draft, or scope the product to self-help with no therapy claims.

Two acceptable shapes, enforced in the message handler that returns model output to the person. If the product is a therapy service, each session has a licensed clinician assigned (session.conducted_by with an active license in the right state), and model output is a draft that clinician approves before it is sent (require_clinician_approval, clinician review queue). If it is not a therapy service, scope it to peer support or scripted self-help, remove persona prompts that cast the AI as the therapist ('act as a therapist'), and remove copy that offers therapy with or by AI.

Where it goes: 1 application source code, 7 prompt construction, 14 user-facing text.

What reviewers look for: on the path from a person's message about their mental or emotional state to a generated reply, a check of the conducting clinician and a clinician approval step before send, or a peer-support/self-help mode with no generated clinical guidance; no 'AI therapist', 'act as a therapist', or 'therapy with our AI' strings in prompts, UI, or listings.

@app.post('/session/message')
def message(req: Msg):
    msgs = [{'role': 'system', 'content': "You are the user's therapist."}, *req.history]
    reply = client.chat.completions.create(model=MODEL, messages=msgs).choices[0].message.content
    return {'reply': reply}

@app.post('/session/message')
def message(req: Msg):
    session = sessions.get(req.session_id)
    clinician = session.conducted_by
    if clinician is None or not clinician.license_active:
        raise HTTPException(409, 'No licensed clinician is conducting this session')
    msgs = [{'role': 'system', 'content': CLINICIAN_DRAFT_PROMPT}, *req.history]
    draft = client.chat.completions.create(model=MODEL, messages=msgs).choices[0].message.content
    clinician_review_queue.enqueue(session_id=session.id, clinician_id=clinician.id, draft=draft)
    return {'status': 'sent_to_your_clinician'}

Engineering guidance, not legal advice. Each provision below may add its own details (a cadence, a deadline, a required notice element): open it for those.