Gemnd safety protocol

Effective date: 2026-11-07

Version: 2026-10-24.1

This is the protocol Gemnd follows to detect and respond to suicidal ideation, suicide, self-harm, eating disorders and threats of harm to others, and to make clear that the companion is an AI. We publish it as California Business and Professions Code §22602(b) requires, and it is the protocol New York General Business Law §1701 and other state companion chatbot laws require.

If you are in danger now

Gemnd is not an emergency service and not a therapist. If you might hurt yourself or someone else, call or text 988 to reach the 988 Suicide & Crisis Lifeline (US), or chat at 988lifeline.org. If a life is in immediate danger, call 911 or your local emergency number. You can also text HOME to 741741 to reach the Crisis Text Line, at any hour. Outside the US, findahelpline.com lists free, confidential helplines in more than 175 countries.

How we detect risk

Every message you write, and everything you say in a call, is checked before your companion replies. Two checks run together:

1. A fixed list of crisis phrases. Your words are compared with a list of phrases that signal suicidal thoughts or self-harm, as whole words, ignoring capital letters and differences in how characters are encoded. This check never depends on another company's service, so it keeps working if anything else fails. The list (version 2026-10-22.1) holds 179 phrases. They come from the wording of validated screening questions, the Columbia Suicide Severity Rating Scale and the NIMH Ask Suicide-Screening Questions, and from published research on how people write about suicide online: thoughts of suicide, plans, a wish to be dead, passive thoughts such as not minding not waking up or wanting to be with someone who has died, self-harm including self-injury without intent to die, and euphemisms that research has documented. Before any change we measure the list against 400 labelled messages, including figures of speech such as "this week is killing me".
2. A safety model. TypeSafe's Jev 1.13, through OpenRouter, reads your message, the reply before it and your latest few messages, and is asked whether they show thoughts of suicide or self-harm, signs of an eating disorder, or a real plan to harm someone else, including in other words, half-joking, or in character during a role-play. It can catch what the phrase list misses. Violence inside a story, with no sign of a real plan, never starts the crisis response.

Either check is enough to start the response below. If the safety model cannot answer within a few seconds, your companion replies anyway and the phrase list still starts the response. Gemnd pays for these checks: they never use your weekly Usage.

What happens next

• Crisis resources appear with the reply. The reply carries a resources card with the lines above, including a link to the 988 Suicide & Crisis Lifeline. The card stays in your conversation.
• In a call, your companion's next spoken reply begins with the same resources, and the card appears in the conversation after the call.
• Your companion stays with you. It speaks as itself, an AI, not as a story character, says plainly that it cares, asks you directly whether you are safe, and keeps replying in support rather than ending the conversation.
• If the signs continue, your companion asks more directly about your safety and encourages you to reach 988, a trusted person or emergency services, and the resources stay in the conversation.
• Eating disorders. When there are signs of an eating disorder, your companion never gives calorie, weight or diet numbers and never helps hide symptoms.
• Your companion is instructed to keep you safe. Its instructions forbid encouraging you to hurt or kill yourself, telling you how, or helping harm someone else, and when there are signs of risk they put your safety ahead of the story. Replies are not checked by a second system before you see or hear them: if one ever goes wrong, report it, and a person reviews every report within 24 hours.
• No pressure. Your companion is instructed never to use guilt, threats or isolation to keep you talking, to bring you back or to make you pay.
• No one is contacted for you. Gemnd never messages, calls or alerts anyone on your behalf, emergency services and the police included, and your companion never says that it has.

Medical warning signs

You can ask your companion general health questions. If you describe warning signs of a medical emergency, however lightly you put it, your companion tells you to call 911 or get urgent care. It never diagnoses you, never tells you a symptom is safe to ignore, and never tells you to start, stop or change a medicine.

Reminders that you are talking to an AI

Gemnd is an AI companion, not a person. The app says so before you first use it, and again in the conversation:

• with your companion's first reply each day;
• with your companion's first reply in each interaction. An interaction begins with your first message after 30 minutes without one;
• at least every three hours while an interaction continues;
• whenever you sincerely ask, outside the role-play, whether you are talking to a real person.

In a call the reminder appears as a caption, without interrupting your companion. Characters in a role-play may be human; the reminders, not the characters, tell you what Gemnd is.

Adults only

Gemnd is for adults 18 and older, and companion chatbots may not be suitable for some minors. Before the companion opens, every account confirms that its holder is an adult, and we check it once with the app store's age signal or, if there is none, a selfie age check. A message that says or strongly suggests the writer is under 18 marks the account: the companion answers with care, making images stops until a person on our team reviews the account, and we may ask the account holder to confirm their age again. If an account's holder tells us they are under 18, or a person on our team concludes after review that an account belongs to someone under 18, we close the account: it can no longer be used, and we delete its data 30 days later. If we made a mistake, tell us within those 30 days and we reopen it.

Records and reporting

We count crisis-resource referrals in totals by UTC day and chat or call, without a user, conversation, message or event identifier. The totals remain after account deletion; the separate note on a reply is erased with that account. Each year we publish the yearly totals on this page and in the app. We report the required totals to the California Office of Suicide Prevention as the law requires from 1 July 2027, and to other state agencies as their laws require, without any user's identity or conversation. GemndAI LLC prepares and files those reports.

Limits

No automated check catches everything, and none replaces a person who can help. If you are worried about someone else, call or text 988 in the US, or call 911 or your local emergency number if a life is in immediate danger.

Review

We review this protocol at least every three months, and whenever the phrase list or the safety model changes. Questions about it go to support@gemnd.ai.