Safety & crisis protocol

Hinvel's characters are written by AI. Some people come to stories on hard days. This page explains, in plain terms, what we do when a message suggests someone might be in danger, and what we don't do. It is our published protocol under California SB 243, New York General Business Law Article 47, Oregon SB 1546 and Washington RCW 19.440.

1. Where it applies

Everywhere you can type your own words to an AI on Hinvel: Talk with characters, Live stories, free input in stories, the rewind hub, the story creation chat, and comments.

2. How we spot a crisis

• Before anything is sent to an AI, every message you type is checked against patterns for suicidal thoughts, self-harm and eating disorders, in English and Chinese. This check runs in your browser and again on our server, and it never switches off.
• On the server, an automated classifier then reads the message for a person expressing intent to hurt themselves. Fiction a character lives through is not treated as a crisis; your own words are.
• Automated checks can miss things or misread them. They are a safety net, not a diagnosis.

3. What happens next

1. The story pauses and your message is not sent: not to the AI, and not posted as a comment.
2. We show people you can contact right now:
   • United States: call or text 988 (988 Suicide & Crisis Lifeline): 988lifeline.org
   • UK & Ireland: Samaritans 116 123: samaritans.org
   • Elsewhere: a free, confidential helpline near you at findahelpline.com
3. If it happens again within a short while, we say so plainly and point you straight to those numbers, without asking you to explain anything.
4. You can go back to the story whenever you're ready. Nothing about you is flagged or punished.

If you or someone else is in immediate danger, contact your local emergency number.

4. What characters never do

• Every AI reply is checked before you see it. A reply that describes methods of self-harm or suicide, or encourages them, is withheld.
• Characters are instructed never to claim to be human. If you ask whether you're talking to a person or an AI, Hinvel tells you directly, outside the story: you're talking with AI.
• Characters don't guilt you into staying, coming back or spending, and they don't act hurt when you leave.

5. Reminders that you're talking with AI

• When you start chatting, a short notice says the characters are AI and not real people.
• After every 3 hours of continued chatting, we remind you again and suggest a break. This reminder can't be turned off. The separate break reminder for reading can be changed in Settings.
• Every chat screen also carries a line saying its content is AI-generated.

6. Minors

Hinvel is for adults (18+) only, and may not be suitable for some minors. We ask for your date of birth before you play. If you tell us you're under 18, you can't use Hinvel.

7. What we record

Each time help resources are shown, we keep one anonymous count: when it happened, which check caught it, and which part of Hinvel it was in. We don't record who it was, what they wrote or which story they were in.

8. Questions

Questions or concerns about this protocol: hello@hinvel.com. More about how AI is used on Hinvel: About AI content.

9. How many times we've shown help

The count below updates on its own. We publish it at least once a year, and it never identifies anyone.