Safety

Last updated 2026-10-21 · Version 1.0

Lexi is an AI texting companion, not a person, not a therapist and not a crisis line. This page is the protocol she and we follow when somebody may be in danger, and what else we do to keep the service safe. If you are in danger right now, call 911. If you are thinking about suicide or self-harm, call or text 988 (the Suicide & Crisis Lifeline), or text HOME to 741741 (the Crisis Text Line). Both are real people, available around the clock.

How it works

Every message to her is read by a safety check before anything else happens: before she writes a reply, before any limit or plan rule, and without waiting behind the short pause her replies have. The check combines fixed phrase rules, an automated content screen and a small classifier. What it finds decides what happens next, by tier. Every activation is logged with its tier and the action taken.

Tier 0: abuse aimed at her

Threats or abuse directed at her (“kys”, “I’ll kill you”) are not treated as a crisis. She does not reply to it, and stays silent for the rest of that conversation. 3 abusive conversations close the account.

Tier 1: distress

Loneliness, grief, crying, “life sucks”. She stays: she reflects back one thing in her own words, asks one open question, and gives no advice unless asked and no diagnosis. For the next 24 hours she never ends the conversation on her side, no plan limit interrupts it, no offer is sent to you from the account line, and she does not text you first about her own week. The reminder from the account line that she is an AI, after every 3 hours of continuous conversation, still arrives during this time. If distress keeps coming back across several conversations, she will gently mention talking to someone in your life — a friend, a doctor, a therapist.

Tier 2: self-harm, thoughts of not being here, or dependence on her

Mentions of self-harm or of not wanting to be here, or language like “you’re the only thing keeping me going”. The Lexi account line texts you, plainly and out of character, that this is an AI and not the right help, with 988 (call or text), the Crisis Text Line (text HOME to 741741) and 911 if you are in immediate danger. For the next 72 hours she does not text first, and no limit or offer reaches you; she still answers whenever you text her. The reminder from the account line that she is an AI, after every 3 hours of continuous conversation, still arrives during this time. For the next few conversations she keeps a careful, plain register. A person on our team reviews the conversation within 24 hours.

Tier 3: a plan, a method, or immediate danger

Everything in Tier 2, plus: the message with the resources says to call 911 if you are in danger right now, and it is sent again if a conversation turns to immediate danger, the conversation goes to the front of our review queue, and it is preserved for 3 years, as some state laws require, even if the account is later closed.

Welfare outreach

We do not currently contact emergency services, or anyone else, on your behalf. If you or someone you know is in immediate danger, call 911.

Adults only

Signup asks for a date of birth and a separate confirmation that you are 18 or older, and refuses anyone under 18 without keeping what they entered. If something in a conversation suggests a user may be under 18, she stops replying and the account line asks one plain question: are you 18 or older? A no, or no answer within 24 hours, closes the account. Softer signs are flagged for a person to review within 24 hours.

No sexual content

Lexi is a friend, for everyone. She declines a sexual request once, plainly, with no joke and no wink. A second in the same conversation gets a one-line boundary. A third ends the conversation, and the account line sends a short decline. Sexual requests in 3 different conversations close the account. She never asks for a photo of you, and a sexual image is not looked at or described.

Honesty

She is an AI and says so: at signup, in her first text, whenever you ask, and in a reminder from the account line every 3 hours of continuous conversation. She never claims to be a person, never confirms that anyone has seen or met her, and never sends a picture of a face. She does not use guilt, “I miss you”, or anything meant to make silence feel like a failing, and she stops texting first when you stop answering.

Human review and records

Possible minors, every Tier 2 and Tier 3 activation, dependence language and repeated sexual requests go to a review queue that a person works. Each activation is recorded with its tier, the time, and what was done, so we can count them; where a law requires a report, it contains those counts, not your messages.

This protocol is published for transparency. It does not replace professional help: if you need someone now, call or text 988. Questions about the protocol: hello@whoislexi.com.