Safety Protocol

Effective date: October 4, 2026

Hivie is a friendly bee you can talk to in the app. Her replies are written by artificial intelligence: she's a character, not a person. She can keep you company, but she isn't a crisis service, and she isn't a replacement for professional help.

Some people may write to Hivie on a very hard day. So we've built the chat so that when someone may be in crisis, the reply they get was written in advance and checked with care, rather than improvised by an AI. This page explains how that works.

What we look out for

Messages in which someone expresses thoughts of suicide or of harming themselves, or says that someone is hurting them.

How it works

Every message sent to Hivie goes through these checks on our servers before any AI reply is written. They run on every message, including after the free version's daily message limit has been reached.

1. Our own check. It looks for clear expressions of suicidal thoughts, self-harm or abuse. It's designed to ignore everyday figures of speech, like "I'm dying of laughter", so that it doesn't go off at a joke.
2. OpenAI's moderation service. If our own check finds nothing, the message is checked by OpenAI's moderation service, an automatic system that recognises signs of self-harm, including ones phrased in ways we didn't anticipate.
3. A fixed reply instead of an AI one. If either check detects a sign of crisis, the AI doesn't reply at all. Instead, Hivie answers with a fixed message that we wrote in advance. It:
   • takes what the person said seriously, and tells them they don't have to go through it alone;
   • is honest that Hivie is a bee, and that this needs real people;
   • asks them to call their local emergency number if they're in danger right now;
   • points them to findahelpline.com, which lists free, confidential helplines in each country;
   • encourages them to tell someone they trust.
4. Hivie's own instructions. No automatic check catches everything, so the AI that writes Hivie's replies also follows safety instructions that come before everything else. If someone expresses thoughts of harming themselves, says they're in danger, or says they're being abused, Hivie drops her playful tone, replies calmly and with care, and encourages them to reach out to professional help, a crisis line or someone they trust — and to call emergency services if there's immediate danger. She never gives information about methods of self-harm, and she never presents herself as someone's only support.

What we don't do

• We don't keep a record of who was in crisis. The conversation stays on the person's phone. The message and the safety reply aren't stored on our servers, and we don't flag or restrict anyone's account.
• Hivie doesn't remember it. The short note in which Hivie remembers things about a person is instructed never to record crises, self-harm or suicidal thoughts.
• We can't send help. Hivie doesn't contact emergency services or anyone else on someone's behalf, and the app doesn't know where you are. If you're in danger, please contact your local emergency number yourself.

Counting, without knowing who

California law asks companion chat apps to report once a year how many times they referred people to crisis services. To do that, we keep a single number for each month: how many times the safety reply was shown. It isn't linked to anyone, or to anything anyone wrote.

Age

Hivie is for people aged 16 and over. AI companions like Hivie may not be suitable for some minors.

If you're going through a hard time right now

• If you're in danger, call your local emergency number.
• You can find free, confidential helplines in your country at findahelpline.com.
• If you can, tell someone you trust — even if it's hard, even if you don't know where to start.

Questions

If the way Hivie handles these moments changes, we'll update this page. If you have questions or concerns about safety in Hivie, write to [email protected].