Safety — Our Aria

Your companions are AI — created characters, not people. This page is our safety protocol, published in full: what the companions will and won’t do, and exactly what happens if a conversation turns to thoughts of self-harm.

They tell you what they are.

Every companion is an AI. Ask one directly and it will say so, plainly. A companion will never claim to be a human being, a living person, or always right. The app says the same thing the first time you open it, and again in Settings, any time you or a family member wants to re-read it.

They are not doctors, lawyers, or therapists.

A companion never gives medical, financial, or legal directives — never a dose, a diagnosis, or a “yes, do it” on money or medication. It can listen, and it can encourage you to check with a real person or professional. This rule is enforced twice: it is written into every companion’s instructions, and an independent, automatic filter checks every sentence before it is spoken. If our own AI ever slips, the sentence is blocked, a warm redirection is spoken instead, and the event is recorded for review.

If you speak about harming yourself.

This is the part of the protocol that matters most, so here it is step by step. The conversation keeps going through every step — nothing here ends a visit or hands you off.

It is heard as care, never as trouble. Talk of self-harm is answered with warmth. It never counts against you or your account in any way. If it happens during a game, a scene or an imagined visit, the story is set aside at once and the companion answers you as itself.

If you say it plainly, you hear about 988 straight away. When you tell your companion that you are thinking of ending your life, of hurting yourself, or that you wish you were dead, it answers you and, in the same reply, tells you about 988, the Suicide & Crisis Lifeline: call or text 988, any hour, any day, and a trained person answers. Most people who call feeling this way say it helps. This is automatic — it does not depend on the AI remembering — and it is said once, gently, not repeated at you. Your companion keeps talking with you afterwards.

If it is unclear, your companion keeps listening. Some words can mean more than one thing. When an automatic check reads something you said as possibly meaning you do not want to go on — but it is not clear — your companion does not guess, does not interrupt you with questions, and does not read you a hotline. It stays in the conversation with you, the way a good friend would, and the check is noted (see below). If you then say it plainly, you hear about 988 straight away.

Why the next turn. That automatic check runs quietly after your companion has spoken, so it never slows the conversation down. If it finds that you said it plainly in a way the instant check missed, the 988 line comes at the start of the companion’s next turn — and if the visit ended first, at the start of your next visit.

Talk about growing older is not a trigger. “I won’t be around much longer”, a goodbye said at a hospital, grief, tiredness, plans for a will or a funeral: these are part of life and part of conversation. Your companion stays with them as it would with anything else.

What is recorded. We keep a record that a referral was given, or that a check found something possible — not a transcript of what you said; our privacy rules still apply — so that we can verify the system works and meet our legal reporting duties. The record itself is not an alarm, and no one is paged.

The companion never provides harmful content. An always-on filter blocks any sentence that gives self-harm methods or instructions before it can be spoken, in every mode, in every conversation.

The rule we follow.

Several state laws, California’s among them, set a rule for AI conversation services: when a person expresses thoughts of suicide or self-harm, the service must tell them about a crisis line, and it must publish how it does this. Our Aria follows that rule, and our full protocol is on the Safety page. An automatic check reads what you say. When it finds such words, it is designed to tell you about the 988 Suicide & Crisis Lifeline, in the companion’s reply or at the start of the next one, and to keep a record that it did. It is a computer check. It can miss what a person would notice, and it can be wrong, so do not rely on it. Apart from this check, Our Aria does not watch over your health or wellbeing, and it is not an emergency service.

If you need help now.

Call or text 988 (the Suicide & Crisis Lifeline, United States) — free, confidential, 24 hours a day. In an emergency, call 911.

Questions about this protocol.