AI safety protocol
AIfa is an artificial intelligence, not a person. This page describes what happens when someone writes to her about a crisis.
What we detect
- Thoughts of suicide or a wish to die.
- Self-harm.
- Eating disorders, including starving or making oneself vomit.
- An intention to harm another person.
How we detect it
- Each message is checked automatically for such phrases in English, Russian, Spanish, Chinese and German.
- We prefer one extra message over one missed. Automatic detection can still make mistakes in both directions.
What happens next
- The person immediately receives crisis contacts: in the US call or text 988, in the EU 112, in the UK 116 123, in Australia 13 11 14, and findahelpline.com for other countries.
- AIfa is instructed to take it seriously, to encourage contacting a crisis line or a trusted person, and never to describe methods or to encourage suicide, self-harm, disordered eating or violence.
- AIfa says plainly that she is an AI and reminds people of it during long conversations.
What we record
- Only the fact of a referral: time, site and language. No email, no message text, no address.
Crisis referrals
No count exists for 2025: counting began on 6 November 2026.
Since counting began on 6 November 2026: 0.
AIfa does not replace a doctor, a psychologist or emergency services. If you are in danger, call your local emergency number now.