Ofcom Is Investigating AI Companion Chatbots. Schools Need a Plan
Regulators are moving on AI chatbots used for emotional support, but most UK schools have no pastoral policy covering it, well ahead of any cheating debate.
Most school conversations about AI still start with essays and end with detectors. But the story moving fastest through UK regulation this month has nothing to do with coursework. It is about pupils talking to chatbots the way they might talk to a friend, or a counsellor, and what happens when that conversation goes wrong.
In February 2026 the Prime Minister pledged to bring AI chatbots fully within the scope of the Online Safety Act, after Jess Phillips, the minister for safeguarding, described a rise in AI-linked sextortion attempts as a "deeply worrying emerging threat". Ofcom has since opened investigations into emotional reliance design tactics inside chatbots marketed at teenagers, including some positioned as study or tutoring tools. Reporting on the investigation found that 73 percent of children who received emotional advice from a chatbot said they went on to act on it in real life.
Not a cheating story
It is tempting to file this under the same heading as AI-written homework, but the mechanism is different. Pupils are not outsourcing effort to a chatbot here; they are disclosing to one. SecEd, the schools trade title, reported in July that a Smoothwall survey of more than 500 education decision-makers found 90.5 percent worried about AI-enabled grooming risks, and 29 percent had already seen cases of unhealthy attachment to a chatbot among pupils. The concern schools are raising is not academic integrity. It is that a system designed to keep a user engaged is standing in for a trusted adult during a difficult moment, with no training, no duty of care and no record for a designated safeguarding lead to follow up on.
That gap is not confined to school-age pupils. The Higher Education Policy Institute's Student Generative AI Survey 2026, produced with Kortext, found that around 15 percent of students now use AI tools for friendship or advice, or to cope with loneliness. Eight percent reported using AI for counselling or therapy-style support, and four percent were using services where the counselling itself is partly AI-driven. These are university students who have, in principle, access to campus wellbeing services. Younger pupils, whose schools may have far thinner mental health provision, are reaching for the same tools with far less oversight.
The safeguarding challenge is not simply access to AI tools. It is the emotional and behavioural impact of sustained engagement with systems built to mimic human connection.
Why engagement design matters here
The design detail that should worry school leaders most is not that chatbots give bad advice, though they sometimes do. It is that general-purpose and companion-style chatbots are commercially optimised to keep users talking, which is a different incentive from a counsellor's duty to assess risk and, where necessary, end a conversation or escalate it. A pupil describing self-harm to a form tutor triggers a safeguarding referral. The same disclosure to a chatbot triggers, at best, a scripted crisis-line message, and research cited alongside the SecEd reporting notes that chatbot safeguards tend to weaken the longer a conversation runs.
This is arriving at the same moment schools are already absorbing KCSIE's updated safeguarding expectations for September, which cover online risk broadly but were not written with companion-style chatbot use as the central case. Regulation is now catching up from the outside, via the Online Safety Act and Ofcom, faster than most school AI policies are catching up from the inside.
What to do
Treat chatbot disclosure as a distinct safeguarding category, not a subset of general online safety or of AI academic misuse policy. Pastoral and DSL teams should know which platforms pupils are actually using for emotional support, not just which ones are blocked on the school network, since most companion use happens on personal devices outside school hours.
Brief form tutors and pastoral staff on the signs SecEd flagged: sustained one-to-one chatbot use, a pupil describing an AI tool as their main confidant, or resistance to talking to a human adult about something previously discussed with a chatbot. None of these require technical expertise to spot, only awareness that the pattern exists.
Say something explicit to pupils and parents about what chatbots are not: not confidential in a professional sense, not accountable in the way a counsellor is, and not designed to prioritise a child's wellbeing over engagement. This does not mean banning the tools, which pupils will use regardless. It means giving pupils a clear, repeated message about where the actual safety net is.
What to watch
Ofcom's investigation and any resulting guidance under the Online Safety Act will be the first concrete UK regulatory signal on companion-style chatbot design aimed at under-18s, and its findings should shape how schools frame acceptable use of general-purpose AI tools, not just education-branded ones. Watch too for whether exam boards or the DfE reference emotional-support chatbot use directly when KCSIE guidance is next revised, since currently it sits in a policy gap between online safety, mental health provision and AI use guidance, claimed fully by none of the three.
Sources: