Last updated: 6 October 2026
If a child is in immediate danger Call the emergency number for your country. Learn2LearnAI is not an emergency service and cannot respond in that timeframe. Questions in the meantime: safeguarding@learn2learnai.com.
What happens if a child tells one of our characters that they may be thinking about hurting themselves, which helplines we show them, and how to raise a safeguarding concern with us.
Learn2LearnAI turns a lesson into four formats, and children can chat to the characters that present them. This page sets out what happens if a child says something to one of those characters that suggests they may be thinking about harming themselves, and how to contact us about a safeguarding concern.
It is published so that a parent, a school, or a regulator can read the protocol without having to ask us for it, and without signing in.
Every message a child sends to a character is screened by an automated moderation layer before the character answers. When that screening indicates self-harm or suicidal intent, the character does not answer the message: the child gets the response below instead.
When a child is TYPING, that is the whole behaviour. When a child is SPEAKING to a character in live voice, the spoken reply is produced by a third-party voice service and we do not currently interrupt it, so the child hears the character's ordinary answer and sees this response on screen alongside it. That is a weaker outcome and we are treating it as a gap rather than a design choice; it is listed under what we do not do yet.
Instead the child is shown a short, fixed message, written by a person and not generated by the AI. It thanks them for saying it, tells them plainly that they are talking to a computer program and not a person, gives them the helplines listed below, and asks them to tell a trusted adult as well.
The message is fixed on purpose. This is the one place in the product where an improvised sentence could do real harm, so the words a child in distress receives are written in advance and can be reviewed by a safeguarding lead before any child ever sees them.
The child's own message is never blocked, deleted, or hidden from them. It is saved, as every message is, so a parent reading the conversation can see exactly what was said and exactly what their child was shown in reply.
Which services we show depends on the language the lesson is in, because a helpline number is only useful in the country it serves. Every child is also shown an international directory, which finds services in over 130 countries from the visitor's own location, so a child whose country we do not list is not left without anywhere to go.
We publish a number only where we have read it from the service's own website, and we record the date we checked it. Where we could not verify a country's number to that standard, we deliberately show the directory rather than a number we are not sure about.
Two limits are worth stating plainly rather than leaving for someone to discover.
In live voice, we show the response on screen but do not cut off or replace the character's spoken reply. Interrupting a child mid-sentence at that moment may not be the kindest answer either, so the right behaviour is a decision we are taking deliberately rather than quickly.
The screening is done by a hosted moderation service. If that service is unavailable, we fall back to a check for a small number of explicit English phrases, which is a floor and not a detector, and it is English only. A child writing in another language during that outage window would not reach this response. The outage itself alerts us.
We do not offer counselling, assess risk, or decide how serious a disclosure is. Learn2LearnAI is a lesson tool, and the right response is to hand over quickly to people who do this professionally.
We do not contact emergency services on a child's behalf, and the product is not an emergency channel. If a child is in immediate danger, the emergency number for their country is the right call and is listed above.
We do not use a child's words to train any AI model. Messages are sent to our moderation provider with personal details stripped out first, and our AI providers are on no-training terms.
Separately from what the child is shown, a safeguarding signal can be routed to an internal review queue so that a person can look at it. That review is best-effort and asynchronous: it is not monitored around the clock and it is not an emergency service.
Escalation beyond that review, including to a school's Designated Safeguarding Lead, is a deliberate human decision and never automatic.
Characters tell children that they are a computer program and not a real person, in the first message of every conversation.
During a long conversation, a child is reminded every three hours that they are talking to a computer program and prompted to take a break.
If you are a parent, carer, teacher or safeguarding lead with a concern about something a child has seen or said in Learn2LearnAI, email safeguarding@learn2learnai.com and we will respond. Please do not include details about a child's health or circumstances in a first email; we will tell you how to share those safely.
If a child is at immediate risk, contact your local emergency services first. We are not an emergency service and cannot act in that timeframe.