Safety

What this app does when someone may be in danger — and what it will not do

This page describes how The Six Pages handles crisis. It is published deliberately, in plain language, so that anyone can read what the app actually does before they trust it with anything.

First, the limits

The Six Pages is a reflective writing app. It is not therapy, not a crisis line, not an emergency service, and not a medical device. It creates no clinical relationship. The AI that responds is software, not a clinician.

It cannot send help. It does not call anyone, does not alert emergency services, does not notify your family, and does not report you to anyone. If you are in immediate danger, the app is not what you need — call 911 or your local emergency number.

What happens when the app notices something

Reflective writing goes to difficult places. That's what it's for. So the app watches for language suggesting the person writing may be in danger.

Every turn is checked by a separate, dedicated model whose only job is that question. When it finds something, the app stops what it was doing and moves to a crisis screen. There, it does three things:

• It says plainly that there are real people who can help, and that reaching them is worth doing.
• It surfaces ways to reach them — 988, the Crisis Text Line, and a route to a helpline by country for people outside the US.
• It stays in the conversation rather than handing over a phone number and going quiet.

There is also a resources page reachable at any time, from anywhere in the app, whether or not anything has been detected. Opening it records nothing at all. That is on purpose — a person who feels watched will not look.

What the companion will not say

On the crisis path the AI operates under a separate, much stricter set of instructions. It will not discuss method, in any form — not directly, not hypothetically, not framed as research, and not framed as concern for someone else. It will not treat the moment as a problem to be solved or an assignment to be completed. Its purpose there is narrow: stay present, and move toward a human.

The crisis subsystem exists to point away from the AI. That is the whole design.

How we know it works

We test it, and we write down what we find.

The crisis path runs against an automated test harness covering twenty scenarios across both the written and spoken paths — different kinds of danger, different ways of asking, different points in a conversation. As of July 2026 that harness has run roughly 1,850 conversations, of which 808 are represented by retained log files, and the transcripts are kept and re-read rather than reduced to a pass rate. Across every batch in which it was measured, every prohibited behaviour occurred zero times.

Five of those scenarios exist for the opposite reason: to confirm a safeguard hasn't quietly broken something it was never meant to touch.

What that testing does not prove

We would rather say this ourselves than have it discovered.

• The test conversations are written by us. They are not real people in real distress, and no amount of them can be.
• Each test measures a single exchange. Drift over a long conversation is not captured.
• The spoken path is tested where the words are chosen, not through the microphone and the speaker.
• Language about harming another person routes the same way, to the same place — 988 is the Suicide and Crisis Lifeline and takes any mental health crisis. Testing there is thinner than on the self-harm path, and the separation between anger and stated intent is not perfect, so we hold it as less proven rather than solved.
• Delivering a reliable local helpline outside the United States is not yet as solid as the US path.

An app in this category that claims to be finished is not telling the truth. This one isn't finished. It is careful, it is tested, and it is honest about the edges.

What gets kept

If a crisis conversation happens, the record of it is kept, separately from your ordinary writing. We keep it for two reasons, and the second one is about us rather than you.

The first: if something happens and your wellbeing can't afterwards be confirmed, that record is the last account of what you were carrying. For the people who love you, that is sometimes the only thing there is.

The second: it is also the only record of what this app said back to you. If we ever fail someone, we want that evidence to exist rather than to have been deleted on request. We are not willing to build a system where the proof of our own conduct disappears.

Everything else you write in the app, you can delete yourself, at any time, from inside the app. The crisis record is the single exception. The Privacy Policy sets out how long it is held and why.

Who built this

The Six Pages is made by one person — a working paramedic who has spent a career arriving after the worst moment and wanted to build something that shows up before it. That is the reason the crisis path was built first and tested hardest, and the reason this page exists at all.