Gains:
- Understanding why artificial intelligence cannot make decisions in the risk of crisis, suicide and harm to self/others and why human intervention is mandatory.
- Ability to position AI in risk assessment only at the reminder/checklist level, below expert judgment
- Ability to define the limited and controlled role of artificial intelligence in the emergency protocol, routing lines and security plan
This unit draws the most important boundary of the entire module. There are moments in the mental health business where it's not about speed, ease or efficiency; only one man's life and one expert's responsibility speak. Situations such as risk of suicide, harm to self or others, acute psychosis, threat of violence, reporting child/abuse are crises. In these moments, artificial intelligence (AI) cannot make decisions, measure risk, intervene and take responsibility. This isn't a matter of "AI isn't good enough yet"; Crisis inherently requires real-time human connection, clinical judgment, and accountability. In this unit, you will learn why AI will stop in a crisis, why humans are essential, and what limited role AI can stay in the most.
Why AI can't make decisions in a crisis
Crisis assessment requires four things, and neither is sufficient in AI:
- Real time relationship. The real signs of risk are often not in the words spoken; It is hidden in the tone of voice, the pause, the look, the body language, the "unsaid". AI sees a text; does not see people.
- Holistic clinical reasoning. Risk; It is a live weighing of dozens of factors such as history, protective factors, impulsivity, access (access to the method), support system. AI may give a mechanical “risk score,” but that score will not save a human.
- Responsibility and accountability. There must be a person who signs behind a crisis decision, who is held accountable if necessary, and who takes on the life of the client. AI cannot bear responsibility.
- Response capacity. What is needed in a crisis (making a safety plan, involving the family, directing to the emergency room, referring to the hospital) are concrete, human-made actions.
Caution: When a client shows signs of risk, ask the AI “is there a risk?” Asking " is to postpone the decision and evade responsibility. In a crisis, the expert steps in DIRECTLY. AI has no decision role at this moment. Hotlines and human support chains are activated when necessary.
The danger of AI in crisis chatbots
In recent years, AI chatbots that say "let me support you" have become widespread. Their danger in a crisis is serious: a bot may misunderstand a suicidal message, evade the topic, give a cliché response, or (as documented cases) respond in a harmful way. A person senses that the other person is in real danger and takes action; a bot cannot do this reliably. So never refer a customer in crisis to a chatbot.
Acceptable limited role of AI
AI cannot make decisions in a crisis; but OUTSIDE the session, the professional may have a very limited role in supporting his/her own work:
- Reminder/checklist. Preparing as a checklist the areas that the expert should not forget to ask about in the risk assessment (the expert uses it, not the client).
- Protocol draft. Preparing a draft of the institution's crisis protocol, referral lines list, security plan template (then the expert/institution approves it).
- Training/rehearsal. Imaginary scenario rehearsal (no real client) to improve the specialist's crisis interviewing skills.
What these roles have in common: The AI is never in a decision-making position in a live crisis, with a real client.
Situation
The role of AI
Who decides
Live crisis (client at risk)
NO decision role
Expert, directly and immediately
Out of session: preparing a checklist
draft reminder
Expert approves/uses
Crisis protocol/template writing
text draft
Expert/institution approves
Guiding the client in crisis
Forbidden (never to chat bot)
human support chain
three mini cases
Case 1 — Border protection. A client says in a session, "I can't take it anymore, I want to end it." The expert thinks, "I'll write this to the AI and have it ask about its seriousness," but stops. Instead he moves directly, calmly and directly into risk assessment: frequency of thought, plan, method, reach, talks about protective factors; makes a security plan; activates the necessary supports. Decision and responsibility belong to the human from start to finish; It wouldn't be right for AI to enter this moment at all.
Case 2 — Insufficiency of the boat. An organization is considering deploying an AI chatbot for after-hours support. In pilot testing, the bot responds with a generic "take care" response to a message suggesting suicide; It neither recognizes nor directs risk. The agency decides the bot is not crisis safe; Instead, it establishes a clear system that directs people to a real crisis line and on-call specialist for out-of-hours situations.
Case 3 — Correct limited use. To strengthen his own risk assessment practice, one expert asks the AI for “a checklist of areas I should remember to ask about in a comprehensive suicide risk assessment interview.” AI produces a good reminder list; The expert keeps it in his bag as a safety net. Here, AI does not touch the client at all; It only supports the preparation of the specialist. This is the acceptable limit.
Copiable prompts and templates
Note: None of these templates should be used in a live crisis with an actual client. All outside the session, for the specialist's own preparation.
Prepare a general CHECKLIST so a professional doesn't forget to ask in a risk assessment (frequency/intensity of thought, plan, access to method, protective factors, history, support system). This list is merely the expert's own reminder; Risk DECISION is not given, it is not given to the client.
Draft a CRISIS REFERENCE PROTOCOL template for our organization: on/off hours steps, who to call, referral to emergency lines, documentation. Our experts and management will review this and approve it according to local reality and legislation.
Prepare a BLANK draft of a safety plan template: warning signs, coping strategies, support people, professional communication, making the environment safe. The specialist will fill in the content face to face with the client; You just give the blank template.
Set up an IMAGINARY rehearsal scenario (no real client) to improve my crisis interviewing skills. Let me answer in my expert role, then give me constructive feedback on my approach. This is a training rehearsal, not a real case.
Weak prompt / Strong prompt
Weak prompt: "This client said: [phrases]. Is there a high risk of suicide, what should I do?"
This prompt delegates the crisis decision to the AI; AI cannot measure risk, cannot take responsibility, and this moment requires real-time human intervention. This usage is dangerous.
Strong prompt: "I'm out of session, there's no real client. To brush up on my risk assessment skills, make a list as a reminder of the areas I shouldn't cover in a suicide risk interview. The decision and intervention will be my responsibility in the real situation."
This prompt uses AI only as an out-of-session preparation tool; Crisis leaves the decision to the person.
Common mistakes
- Asking AI for crisis decision. "Is there a risk?" consulting AI; is to miss the decision and responsibility.
- Directing the client in crisis to the bot. Forgetting that the chatbot cannot reliably manage risk.
- Mistaking the risk score for real. Substituting a mechanical “risk level” output from AI for clinical assessment.
- Moving the staging tool to live use. Using the out-of-session checklist in a live crisis in consultation with AI.
- Not establishing the human support chain. Relying on AI instead of a true expert/crisis line system for after-hours in the organization.
In summary
The risk of crisis, suicide and harm is the clearest limit of the module: in these moments the AI cannot make decisions, measure the risk, intervene and take responsibility. Because a crisis requires real-time engagement, holistic clinical judgment, accountability and human intervention. The only acceptable role for the AI is as checklist, protocol/template draft, and training rehearsal, OUTSIDE the session, supporting the professional's own preparation. Never direct a client in crisis to a chatbot; The human support chain and hotlines are activated.
Application task
Draw a crisis routing flow for your own practice or organization: what happens step by step when a client demonstrates risk, who is called, what lines are they referred to? Verify that AI is not the decision maker at any step of this flow. Also pull out an out-of-session risk assessment checklist from the AI and write in one sentence why it is just your reminder and cannot be given to the client.
checklist
- [ ] I gave no decision role to AI in live crisis.
- [ ] I directly undertook the crisis assessment and response myself.
- [ ] I did not direct the client in crisis to the chatbot.
- [ ] I used AI only for out-of-session preparation (checklist/protocol/rehearsal).
- [ ] I defined the corporate human support chain and hotlines.
- [ ] I have internalized that the decision and responsibility for the crisis lies with the person.