Crisis and Self-Harm Safety Protocol

Last updated: October 2026

If you need help right now

You don't have to go through this alone. Free, private help is available 24 hours a day.

R1N.ai is not a crisis service. An AI can't help in an emergency. A real person at one of the services above can.

What this page covers

R1N.ai is an app for adults 18 and over. You talk out loud with AI characters. This page explains what happens when a conversation shows signs that someone may be thinking about suicide, self-harm, or an eating disorder. It also explains what we keep, what we report, and how to reach us.

You are always talking to an AI

Before every conversation, you confirm: “I understand that I'm chatting with an AI, not a person.” During a conversation, a reminder appears every 3 hours. The AI characters are not people. They are not therapists, doctors, or counselors, and they do not provide mental health care.

How we check conversations

Your voice is turned into text as you speak. Every message you say, and every reply an AI character gives, is checked automatically while the conversation is happening. The check is done by a text classifier from an outside company: a computer program that sorts text into categories. It looks for signs of:

• suicidal thoughts or plans
• suicide
• self-harm
• eating disorder behavior, such as starving or purging

It also checks for a few other safety problems, described in our content rules.

AI replies are checked as they are spoken, so the first words of a reply may be heard before the check finishes. If a reply is flagged, it is cut off right away.

No person listens to your conversations. We do not store audio.

How we tested the checks

Before using the checks, we tested them on a public dataset that is used to test content moderation, and on our own set of examples. Some examples were real warning signs. Others were everyday phrases that sound similar but aren't, like “this traffic is killing me.” We chose the settings based on those results. The goal is to catch real warning signs without stopping conversations over ordinary phrases.

No automatic check is perfect. It can miss things, and it can sometimes stop a conversation that was fine. If that happens to you, please tell us using the contact details below.

What happens if something you say is flagged

1. The conversation stops right away. The AI does not reply to what you said. We do this on purpose: an AI character is not the right one to answer you in that moment.
2. A message appears on your screen, with a short sound, listing the 988 Suicide & Crisis Lifeline (call, text, or chat), Crisis Text Line, help outside the US, and 911 for emergencies.
3. Your account is put on hold. This is not a punishment, and you are not in trouble. It gives a real person a chance to check in before the app continues.
4. I review what happened, and I reach out to you. R1N.ai is run by one person, and I do every review myself. I aim to review within 48 hours, and I will email you at the address on your account.

While your account is on hold, please use the services at the top of this page. They are free and open all the time.

What happens if the AI says something harmful

An AI can sometimes say something it shouldn't, such as something that encourages self-harm. Every AI reply is checked too. If an AI reply is flagged:

1. The conversation stops.
2. The same help message appears on your screen.
3. Your account is put on hold until I've reviewed what happened, which I aim to do within 48 hours. You didn't do anything wrong, and I'll email you when it's resolved.

What the AI is not asked to do

We do not rely on the AI to handle a crisis. The AI characters are not given special instructions for these moments. Instead, the system stops the conversation and shows you real places to get help.

Records we keep

When a moment is flagged, we save the few lines of conversation around it, about 6 back-and-forth turns, so I can review what happened. These lines are encrypted, which means they are scrambled so others can't read them. They are deleted automatically after 90 days. We do not store audio, and we do not keep the rest of your conversations for this purpose.

What we report

Each time the help message is shown, we add one to a count. The count does not include your name or any other personal details.

California law requires us to send a yearly report to the state's Office of Suicide Prevention. It includes how many times we showed the help message and how this protocol works, and no names or personal information. We also publish the count here every year.

Loading…

The protocol described on this page is the protocol we used during that year.

Age checks

R1N.ai is for adults 18 and over only. AI companion apps may not be suitable for some minors. Every new user must pass an age check when they sign up. The check uses a selfie to estimate age, through an outside verification company. If the selfie check can't confirm your age, you must show an ID. If a conversation suggests a user may be under 18, it ends and the user must pass the age check again.

Worried about someone else?

If you think someone you know may be thinking about suicide, you can call or text 988 for advice on how to help them. If they are in danger right now, call 911.

Contact and questions

If you have questions about this page, think a conversation was stopped by mistake, or want to report a safety problem, write to me at [email protected]. I read every message myself.

If you are in crisis, please don't wait for my reply. Call or text 988, or call 911.