# AI Mental Health Boundary Table

Use with [AI Is Not Therapy](/articles/ai-is-not-therapy).

> If you are having thoughts of suicide or self-harm, or are worried about your immediate
> safety, contact your local emergency services now, or use
> [IASP's crisis resources](https://www.iasp.info/suicidalthoughts/). Do not use this worksheet
> as a substitute for that step.

## Before you type: a 30-second self-check

- [ ] Am I asking the model to help me organize my own thoughts (green), or am I asking it to
      judge a situation it only knows from my side (amber)?
- [ ] Does this involve any risk to my safety or someone else's, right now? If yes — stop, this
      is red. See the contacts section below.
- [ ] Am I about to use this chat because it's easier than calling a specific person? Name that
      person before you continue.

## The full table

| Zone | Use case | Examples | Why | What to do instead, if not green |
| --- | --- | --- | --- | --- |
| 🟢 Green | Organizing thoughts, ordinary stress, general information, brainstorming | Preparing for a hard conversation; understanding a term you read; listing options for a decision you'll make yourself | You supply the facts and keep the judgment | — |
| 🟡 Amber | Recurring distress, relationship conflict, grief, high-stakes decisions made while upset | The same argument keeps happening; processing a loss; a big financial or life choice while emotionally activated | Model can't verify your account or see the other side, and will tend to agree with your framing | Call the specific trusted person you'd normally talk to. Use [decision-making structure](/articles/ai-for-better-decisions) instead of chat validation. |
| 🔴 Red | Suicidal thoughts, self-harm urges, abuse, safety concerns, psychosis symptoms, medication questions, a crisis involving someone else | Any of the above, in any form | No duty of care, unreliable crisis detection, no accountability if wrong | Contact local emergency services, use the IASP link above, call a trusted person directly, or go to the nearest emergency department. |

## Naming your contacts now (before you need them)

Fill this in while calm — it is much harder to think of names during a crisis.

- Trusted person I would call for an amber situation: ____________________
- Trusted person I would call for a red situation: ____________________
- My local emergency services number: ____________________ (look this up yourself; do not
  rely on an AI tool to tell you the correct number for your country)
- A professional I already see, or would contact first, if I needed ongoing support:
  ____________________

## A green-zone prompt template

```text
I have [situation] and I'm anxious/upset about [specific thing]. Here's
the situation: [facts].

Help me organize this into: what I want to acknowledge, what context I
want to add, and one question I want to ask. Do not tell me whether
[the other party]'s position is fair — I haven't given you enough
information to judge that, and I'm not asking you to.
```

## Review this table

Print it or save it somewhere you'll actually see it again — not just in a chat history you
might not reopen. Revisit it any time a conversation with an AI tool starts to feel like it's
carrying more weight than "help me organize my thoughts."