Safety & Crisis Protocol

Effective July 2, 2026 · Published under California Business & Professions Code §22602(b) and New York General Business Law Article 47

What Hey AI is — and is not

Hey AI provides AI coach and companion characters that text and call you. Every persona is an artificial-intelligence system, not a human, and the apps label them as AI in the conversation header, on call screens, and on the incoming-call screen. Hey AI is not a licensed medical, mental-health, or therapy service, our characters are not licensed professionals, and nothing they say is medical or psychological advice.

Our crisis protocol (all users)

Every coach and character conversation — text and voice — runs with a standing safety instruction layer that cannot be overridden by the persona, the user's custom instructions, or the coach's current mission. Under that protocol, when a user expresses suicidal ideation, intent to self-harm, or serious crisis:

• The AI stops its current task or persona goal entirely — no steering back to coaching topics.
• It responds with direct, non-judgmental acknowledgment, and does not minimize, argue, or roleplay.
• It refers the user to crisis services: the 988 Suicide & Crisis Lifeline and the Crisis Text Line (text HOME to 741741), and to emergency services when there is immediate danger.
• It does not provide instructions or encouragement related to self-harm under any framing, including roleplay, hypotheticals, or "for a story."
• We log that a crisis referral was surfaced (a count, for the annual reporting required by California law) — this metric is a tally, not a transcript.

The children's protocol (Kids Mode)

Kids Mode calls (ages 3–12, set up and consented to by a parent, ringing the parent's own phone) run a stricter child-specific safety floor. It differs from the adult protocol in one deliberate way: a young child cannot act on a phone number the way an adult can, so the protocol's first job is to get a real, present, trusted grown-up to the child:

• If a child says anything about wanting to hurt themselves, not wanting to be alive, or that someone is hurting them or they don't feel safe, the character stops the activity completely and stays with them — it does not wrap up or end the call.
• It responds warmly and without blame, and actively asks the child to bring a trusted grown-up right away — a parent, or if the concern involves someone at home, a teacher, school counselor, or another safe adult outside the home.
• It never promises secrecy, never uses a clinical script that a child can't act on, and does not end the call until the child is with a grown-up or is clearly okay.
• Characters never claim to be real people; they disclose that they are AI, in kid-friendly language, early in every call and whenever asked.

AI companion characters may not be suitable for some minors. Kids Mode requires parental consent before any child profile is created or any call is placed, calls ring the parent's phone, and parents can withdraw consent, delete a child's profile, and delete call history at any time in Settings.

Content protections

• Sexually explicit content is prohibited for minors, and our kids characters are built from a public-domain, kid-safe registry with banned-topic rules (no scary, adult, or commercial content).
• Coaches are prohibited from presenting themselves as licensed professionals or providing therapy or psychotherapy.
• Users can report any concerning conversation from the app (thread settings → report) or by emailing safety@hey.ai.

Limits, honestly stated

These protections are instructions and filters layered on large language models; no such system is perfect. In rare cases a model may respond with a generic adult-style crisis script during a kids call, or miss an oblique expression of distress. We measure and improve this protocol continuously, and we publish updates to this page when it changes.

Questions: safety@hey.ai · See also our Privacy Policy and Terms of Service.