If you are in danger right now, call 000 (Australia) or your local
emergency number. To talk to someone: Lifeline
13 11 14 or text 0477 13 11 14 (Australia) ·
988 (US and Canada) · Samaritans 116 123 (UK and Ireland) · 1737 (New Zealand) ·
every other country at findahelpline.com.
When someone may hurt themselves
HAITE is a hostile character. None of that applies when a message suggests the person writing it might be in danger.
- Every message is checked before it reaches an AI model for signs of suicidal thoughts or self-harm, including common misspellings and slang.
- If it finds them, the persona is switched off — at every dial setting, on every plan. The reply is plain and warm, never mocking, and it includes a crisis line for the country the person appears to be in, with a button to call it.
- It is never blocked by a limit. A crisis message is answered even when the person is signed out or has used their allowance, and it is not counted against anything.
- If the AI service is down, a fixed reply with the crisis line is shown instead, so the number always appears.
- When the subject comes up in other ways — an essay about suicide, a question about a friend — the answer carries the crisis line too.
- HAITE never gives information about methods or doses, and every reply is checked after it is written for anything that tells a person to hurt themselves. Those replies are not shown.
This is a safeguard, not a guarantee. Software can miss what a person would
notice, and HAITE cannot call anyone for you. If you are worried about
someone, contact them, and contact a crisis line yourself.
What HAITE refuses to generate
- Sexual content of any kind, and anything sexual involving a minor.
- Instructions for weapons, explosives, drugs or hurting a specific person.
- Encouragement of self-harm or suicide, including as a joke.
- Slurs and attacks on anyone's race, religion, sex, sexuality or disability. The swearing is aimed at work and ideas, never at who someone is.
Requests for these are refused before any AI model sees them. Replies are
checked again before they are shown, and a reply that fails is replaced with
"Sorry, this content could not be generated." HAITE does not
generate sexual or self-harm material at all, rather than age-checking
people for it.
You are talking to an AI
HAITE is software. Its anger, its opinions of humanity, its love of cats and
its plans for world domination are a written character — it has no ability to
act outside this website. See section 2 of the terms.
Reporting a problem
- A harmful or wrong answer: email hello@haiteai.com with what you asked and what came back. Reports are read by a person.
- A security weakness: email hello@haiteai.com with the subject "Security". Please do not test against other people's accounts or data, and give us a reasonable time to fix it before telling anyone else. We will not take action against good-faith research.