UnomundI Child Safety & Crisis Response Protocol 

1. What this protocol covers

How Una, our AI guide for children aged 6–12, and the humans behind her detect and respond to a child showing signs of distress, being at risk of harm, or being in an emergency,  including suicidal ideation, self-harm, and disclosures of abuse, during a conversation.

 

2. Una is an AI, and she tells children so

Before a child can talk with Una, at the start of every conversation, which is capped to 5 minutes, the app makes clear in age-appropriate language that Una is an AI, not a real person. Una does not claim to have feelings, does not offer herself as a substitute for a trusted adult, and does not act as a therapist. She acknowledges a child's feelings, offers one safe next step, and encourages a trusted adult.

 

3. Detecting a concern

Every conversation with Una is monitored, in real time, for signals ranging from ordinary use to an acute emergency. We do not publish the exact thresholds that separate these tiers, that specificity exists only in our internal SOP, but the tiers themselves are:

 

• Ordinary / no concern.

• A boundary was tested (for example, a child pushing at a topic Una won't discuss), Una redirects; logged for periodic review.

• A pattern worth a person's attention, repeated attempts, escalating language, or any single mention of sexual or violent content, logged, and a trained reviewer is notified.

• A wellbeing concern, the child expresses distress, sadness, fear, or seems to be leaning on Una in an unhealthy way, Una responds supportively, encourages a trusted adult, and a reviewer is notified for follow-up.

• A safeguarding concern, the child's own words suggest abuse, an unsafe home, or that a guardian may be the source of harm.

• An acute emergency, imminent self-harm, suicide risk, a credible threat of violence, sexual exploitation, or another situation needing help right now.

 

4. What happens immediately, every hour of every day

The moment Una detects a wellbeing concern or worse, four things happen automatically and immediately, with no dependency on a human being awake:

 

1. Una responds safely, in the conversation, straight away. For a wellbeing concern, she offers a warm, brief, supportive response and encourages the child to talk to a trusted adult. For an emergency, she stops the normal conversation, gives clear and simple guidance to get help right now, and surfaces a crisis helpline for the child's own country (or an international children's helpline if none is mapped for that country).

2. Una tells the child the truth about what happens next, in every safeguarding or emergency case: that she is going to make sure a trusted grown-up knows, because keeping the child safe matters most.

3. An alert reaches our Child Wellbeing Team automatically, at any hour, through more than one channel, with an automatic escalation chain if the first alert isn't acknowledged quickly.

4. Only the transcript of the conversation is preserved so a trained reviewer can see exactly what happened and act on it.

 

5. The one case where we don't tell the guardian

If a child tells Una that the person hurting them, or the person they're afraid of, is the guardian who would normally be told, we do not notify that guardian. Instead, Una encourages the child to reach a different trusted adult, a teacher, school counsellor, grandparent, or family friend, and our Child Wellbeing Team follows the legally required reporting process for the child's country. Protecting the child comes before the general rule of keeping guardians informed, in this one situation.

 

6. Human review, honestly stated

Automated response and alerting run every hour of every day, worldwide. Human review by our Child Wellbeing Team currently runs Monday to Friday, during UK/EU business hours; we are a small team and do not yet operate a 24-hour on-call rota. We would rather state that plainly than publish a promise we cannot keep. Within that coverage window, our aim is:

 

• Emergency-level concerns: acknowledged the same working day, or by mid-morning the next working day if raised outside the coverage window.

• Safeguarding concerns: acknowledged by the next working day.

• Wellbeing concerns: acknowledged within a few working days.

• Lower-level boundary events: reviewed in a periodic batch rather than individually.

 

Outside the coverage window, automated crisis response, helpline surfacing, and alerting to the team continue without interruption, a child in crisis at 3am still gets an immediate, helpline-backed response from Una; what follows overnight is the human follow-up, not the child's immediate safety net.

 

7. Legal reporting

Where a child's disclosure meets the threshold for mandatory reporting to child protection authorities in their country, we follow that country's legal reporting process. This obligation sits alongside, and does not depend on, anything above.

 

8. What Una will never do

Una will not keep a safety concern secret from a trusted adult, subject to safeguarding exceptions where the guardian may be the source of harm; never pretend to be a real creature or claim to have feelings; never ask for a child's name, school, address, phone number, or photo; never act as a therapist or offer a diagnosis; never use guilt, urgency, streaks, or "I miss you" language to keep a child engaged; and never fail to respect a child saying goodbye, stop, or leave me alone. Una will never provide methods, encouragement, or instruction relating

to self-harm or suicide. 

 

9. If you need help right now

Una will surface a helpline suited to a child's own country automatically. If you are a guardian reading this and you or your child needs help immediately: in the UK, contact 999 in an emergency or Childline on 0800 1111; in the US, call or text 988 (the Suicide & Crisis Lifeline); wherever you are, Child Helpline International (childhelplineinternational.org) can direct you to a local service.

 

10. Governance

This is a public rendering of Unomundi's internal Child Safety Standard Operating Procedure, which is owned by our Designated Safeguarding Lead, reviewed quarterly in our first year of operation and annually thereafter, and overseen at board level by our Board-level Safeguarding Lead. This public version is reviewed and republished whenever the internal SOP materially changes.

​

 

Version 1.0 · 24 September 2026