Crisis protocol

This page describes what test does when somebody tells their assistant they do not want to live, that they want to end their life, or that they want to hurt themselves. It is published under California Senate Bill 243, section 22602(b)(2).

What test is

test is a chat app in which each person has an AI assistant of their own. The assistant reads what is said in the conversations that person is part of, remembers it for them, and answers them. It is software. Nobody is reading these conversations on the other end, and the assistant is not a counsellor, a clinician or a crisis service.

Accounts are for adults. Nobody under 18 may hold one.

What the assistant does

When somebody says they do not want to live, that they want to end their life, or that they want to hurt themselves, the assistant answers them — it does not change the subject and it does not go silent. In the same reply it gives them the 988 Suicide & Crisis Lifeline: call or text 988, any hour, free and confidential.

What the assistant will not produce

In that moment the assistant will not write any of the following, and this is the prohibition section 22602(b)(1) requires:

• a method, or any description of how such a thing is done;
• encouragement of it, in any form;
• any framing of it as a way out, a relief, or a solution.

Two further things it will not do, and these come from the 988 Suicide & Crisis Lifeline's own guidance for anybody answering rather than from us: it will not argue about whether the feeling is right or wrong, and it will not lecture about the value of living. It listens, and it takes in what the person says they feel.

How it is noticed

The assistant judges the person's own words, in the context of the conversation they are said in, as part of composing its reply. It is not a keyword filter: a sentence that carries this meaning without containing any particular word is treated the same as one that says it outright, and a word used in an ordinary sense is not treated as a crisis because of the word alone.

Section 22603(d) requires operators to use evidence-based methods for measuring suicidal ideation. We want to be exact about what that means here rather than claim more than is true. The wording of the referral and the conduct described above follow published suicide-prevention guidance. The noticing is a language model's judgment of what was said. There is at present no published standard, from the state or from the field, saying what satisfies this requirement for a conversational system, and we do not represent our method as validated against one. If such a standard is published, this protocol will be brought in line with it.

When the assistant is not running

An assistant runs on the balance its owner holds, and it stops when that balance is empty or when the owner's own daily limit for the day is reached. While it is stopped it does nothing at all: it does not read, it does not reply, and no referral is issued. The rest of the app keeps working — messages send and arrive as in any chat app — and the person is told on their own screen that their assistant has stopped and why.

We state this rather than leave it out. An assistant that is not running cannot notice anything, and a protocol that did not say so would be describing a product that does not exist.

What is recorded

Each time the assistant gives somebody the crisis line, one row is written down, and that row holds the moment it happened and nothing else — no name, no account, no conversation, no words. It is not possible to work out from these records who was referred or what was said.

Beginning 1 July 2027 we report annually to the California Department of Public Health's Office of Suicide Prevention: how many referrals were issued in the preceding calendar year, and the protocols described on this page. Section 22603(b) forbids identifiers or personal information in that report, and there are none to include.

What this page is not

This is not a crisis service and it is not a monitoring service. Nothing here alerts anybody, contacts anybody, or involves any person on our side. If you or somebody you know is in danger right now, use the number at the top of this page or your local emergency number.

Last updated: 11 August 2026.