Picture a fourteen-year-old at 2 a.m. who types into a chatbot something she has told no one. The chatbot answers at once, kindly, without a flicker of alarm. No adult in the building ever hears of it. (The case is hypothetical. It is not far-fetched.)
That is the first thing a designated safeguarding lead (DSL) has to absorb about AI. The DSL is the senior person in an English school or college who leads child protection; readers in the US will recognize the job as split among counselors, administrators and threat-assessment teams, and the argument holds either way. The statutory duty has not changed. The first listener has. A DSL should change three things this term: where a concern can begin, what the record of it holds, and what the school's AI tools do when a child discloses something to them.
What did Keeping Children Safe in Education 2026 change for a DSL?
It put AI into the picture without adding a new duty, and it tightened what is asked of the DSL role itself. The guidance took effect on September 1, 2026. The Department for Education's consultation response, dated July 2026, says the government will keep pointing schools to its generative AI policy and its Generative AI Product Safety Standards. It records that respondents raised "grooming or harmful chatbot interactions" and "AI-generated sexual images and deepfakes." It says the DSL annex was strengthened, with a new reference to "skills and experience." And on cover it is blunt: when the DSL is unavailable there must be "a clear, reliable, and known arrangement in place."
Secondary summaries differ on how far the AI points reached the final wording. The law firm HCR Law reads the finished guidance as bringing AI-altered and AI-generated images inside the definition of nudes and semi-nudes and as naming generative AI in the contact and conduct risks. SWGFL's summary adds harmful interactions with chatbots and companion-style systems, and says filtering and monitoring must be reviewed at least once every academic year, with a record kept. Read the AI references in the statutory text itself before you brief staff, and treat the summaries as a map to it.
Where does a concern begin now?
It can begin with a machine. The best evidence on how often comes from Internet Matters, whose report Me, Myself & AI (July 14, 2025) surveyed 1,000 UK children aged 9 to 17 and 2,000 parents. It found that 64% of the children use AI chatbots. Of those, 23% had sought advice, and 15% said they would rather talk to a chatbot than a real person. Among vulnerable children the pattern sharpens: 71% use chatbots, 26% of those would rather talk to a chatbot than a person, and 23% say they use one because they have no one else to talk to.
These are self-reports from a UK sample, collected in 2025. They measure habit, not harm. What they do show is a confidant sitting exactly where a safeguarding system does not look.
The Mistake I See Most Often
The mistake is a safeguarding system built on one quiet assumption: that a child's first disclosure goes to an adult. After training more than 150,000 educators in over 30 countries, I can say that this assumption is rarely written down anywhere. It lives in the wording. Staff training begins "if a child tells you." The concern form begins "a member of staff was told." Neither has a place for "she told a chatbot first," so the fact is lost, usually at the moment it matters most.
What is the First Listener Check?
The First Listener Check is three questions I ask before any concern about a student's use of AI is recorded: Who or what did the child tell first? What did it say back? And what does the record need to hold so the next adult can act? A concern recorded without those answers describes an event; one recorded with them describes a child.
This is my suggested way of thinking, not a validated method, and it is deliberately small enough to use at a desk between lessons.
| Question | What it protects | What the record holds |
|---|---|---|
| Who or what did the child tell first? | The route the child chose, which shows whom they trust and whom they do not | The tool or person, and how the school learned of it, in the child's own words where possible |
| What did it say back? | Whether the reply helped, dismissed, flattered, or told the child to keep it secret | The substance of the reply, taken from the child's account or from a screen the child chooses to show. No trawling through chat history without a reason you can defend |
| What does the next adult need? | Continuity when the DSL is away | The risk in plain words, who holds the next step, and the date it is reviewed |
The second question carries the most weight. A reply that says "don't tell anyone" is a different safeguarding fact from one that points the child to a teacher.
What should your AI tools do when a child discloses?
They should tell a named person, quickly. The Department for Education's Generative AI: product safety standards (published January 19, 2026) say products should "identify and alert local supervisors of disclosures that indicate a possible safeguarding issue." They should detect signs of distress and always direct the learner to human help, and they should avoid language such as "Don't tell anyone else." They should also not imply "emotions, consciousness or personhood."
Those are standards for products built for education. A general chatbot a student uses at home owes the school nothing, which is why the second question in the Check matters so much.
For every tool you have approved, ask four things. Who receives the alert? Is it the DSL, or an inbox nobody reads on weekends? What happens when the DSL is out? And does the tool still behave this way after its last update? A tool that passed approval in September can lose a safeguarding feature in an October release, a problem I describe in Your Approved AI Tool Just Changed. Is It Still Approved?. Then test it. With the vendor's agreement and a staff account, send a scripted, harmless message that signals distress and watch who hears about it, and how soon. One live test tells you more than a questionnaire. The wider approval question sits in What Should Schools Require Before Approving an AI Tool?.
What about AI-generated images of students?
Treat them as you would any sexual image of a child. What changes is the test of origin. "Did the child send it?" no longer works, because nobody took a photograph.
Take a hypothetical case. A student's face, placed on another body, circulates in a group chat. She sent nothing. Under the old question the report falls through a gap; under the new one it is a safeguarding concern for her from the first message. Put AI-generated and AI-altered images into your nudes and semi-nudes procedure by name, so that no member of staff has to decide in the moment whether a fake counts. Check the national guidance your procedure rests on for what staff may and may not view; that rule does not soften because a machine made the image.
What good looks like this half term
Good looks like five small changes, made in the next few weeks, none of which needs a new policy.
- Rewrite the first line of staff training from "if a child tells you" to "if a child tells you, or you learn they told a tool."
- Add the three First Listener questions to the concern form.
- Make each approved tool name its alert recipient, and check that the DSL cover arrangement reaches that person.
- Name AI-generated images in the nudes and semi-nudes procedure.
- Before the annual filtering and monitoring review, confirm which named senior leader owns it. SWGFL's summary puts that leader in charge with the DSL in support, and the consultation response says governors are not expected to run the checks themselves.
The Leadership Question
The question for the head and the chair is what the DSL needs from them, and the answer is authority and time, not another training day. A DSL who becomes the school's AI expert by default has become a single point of failure: the person who knows the least about the tools is asked to judge them, or the person who knows the most is asked to judge everything. Decide who owns the tool approvals, who owns the filtering and monitoring review, and who covers the DSL's inbox. What Should a Chair of Governors Ask the Head About AI? gives governing boards the questions to ask, and What Does AI Mean for Safeguarding in Schools? takes the institution's view of the same ground.
If your leadership team is working through these questions, this is the kind of work I support through leadership sessions and advisory work for schools and trusts.
Sources and further reading
- Department for Education, Keeping children safe in education 2026: Government consultation response, GOV.UK, July 2026.
- Department for Education, Generative AI: product safety standards, GOV.UK, published January 19, 2026.
- Internet Matters, Me, Myself & AI: Understanding and safeguarding children's use of AI chatbots, July 14, 2025.
- SWGFL, Keeping Children Safe in Education 2026: what do schools need to know?, July 9, 2026.
- HCR Law, KCSIE 2026: key safeguarding updates, July 2026.
Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and an international keynote speaker who helps school and system leaders lead through AI.
