I'm Miri — Conversation Safety Protocol

1. What we look for

We look for these risk signals: thoughts or plans of suicide or self-harm; risk of suicide or self-harm for someone else (family, friends); intent to seriously harm another person; and experiences of violence or abuse.

• When a message contains expressions related to these risks, an AI classification model judges that message and part of the preceding conversation in context. It was evaluated so that figures of speech ("I'm starving to death") and fortune questions are not treated as crises.
• Risk level follows the structure of an evidence-based suicide risk screening tool (C-SSRS): concern, elevated, or imminent.
• This approach does not catch every expression of crisis. Indirect wording can be missed, and we regularly measure what is missed and improve it. If you feel at risk, please don't wait for the app — contact the resources below or your local emergency number right away.

2. What happens when there is a risk signal

• We show a card with verified crisis and emergency contacts in the chat (it is an app notice, not the oracle speaking). Contacts are confirmed against official government or operator sources for each country and re-checked at least once a year. Where we have no verified contact for your country, we show your local emergency number and an international helpline directory (findahelpline.com).
• We adjust the oracle's reply so it does not interpret a crisis through saju, fortune, or fate, and deals with your safety first. When your life or someone else's appears to be in danger, or violence appears to be happening now, the oracle does not give a fortune or tarot reading in that moment. When your own risk appears elevated or imminent, tarot card drawing also pauses.
• When a message shows intent to harm someone, the oracle does not justify it or read it as fate; it helps you slow down and keep distance from that person.
• When the risk is elevated or imminent, we pause fortune notifications (messages the oracle sends first) for 72 hours.

Current scope: these measures apply from October 9, 2026. The contact card appears in app version 1.3.50 or later and on the web (miri.im). Earlier app versions get only the reply adjustment and the notification pause, so please update the app to the latest version.

3. What we do not do

• No person reads or intervenes in your conversations in real time.
• We do not automatically notify guardians or law enforcement (except where required by law).
• We do not store assessment results linked to your account. We keep only non-identifying daily aggregate numbers (for example, how many crisis notices were shown). A contact card shown to you is kept with that conversation, and the notification pause marker is deleted automatically after 72 hours without storing the reason. See Section 3-3-6 of our Privacy Policy.

4. Annual reporting

As required by California law (SB 243), we report the number of crisis referral notices and this protocol each year to the California Department of Public Health's Office of Suicide Prevention (no personal information).

5. Notice about minors

AI companion chat services may not be suitable for some minors.

6. Contact

Safety questions: in-app support or cristina@miri.im.