The Unseen Syllabus: Pupil Scrutiny and the Governance of School AI

Artificial intelligence is arriving in UK schools less through a debate about pedagogy than through a procurement portal. A trust signs a licence for an adaptive tutoring platform; a headteacher trials an AI lesson-planning assistant; a local authority or multi-academy trust renews a analytics package that claims to predict attainment. By the time pupils encounter the system, the important choices have often been made: what data is collected, how outputs are used, whether a human reviews a decision, and what happens when the tool is wrong.

That is a governance problem before it is a technical one. Democratic education insists that people affected by decisions should have a meaningful say in them. Pupils are not merely users of school AI; they are its subjects. Their work, behaviour, attendance and sometimes their inferred attitudes become inputs. A school that teaches citizenship while excluding pupils from the governance of these systems is teaching a quiet lesson of its own.

Why pupils belong in the conversation

Pupil voice in AI is not a matter of asking children to approve code. It is a matter of legitimacy, scrutiny and practical wisdom. Pupils notice things that procurement teams miss: whether a chatbot responds differently to a shy learner, whether a dashboard encourages narrowing to predicted grades, whether a revision tool rewards confidence over understanding. They also bear the consequences of error. An algorithm that misreads a pupil's needs is not an abstract risk; it can shape sets, interventions and self-belief.

There is a second argument, more institutional. Schools are under pressure to demonstrate that they take pupil voice seriously. Too often that evidence is a survey, a school council minutes file or a poster campaign. AI governance offers a sharper test. It involves contracts, data protection, assessment and spending. If pupil voice can reach these questions, it is not decoration. If it cannot, governors should ask why.

The limits of the present arrangement

Most schools and trusts are not acting in bad faith. They are acting under capacity constraints. Governors may receive a briefing after a product has been purchased. Data protection impact assessments, where they exist, are dense and rarely summarised for a lay audience. Vendors present polished demonstrations. Teachers are understandably drawn to tools that reduce workload. Pupils are consulted, if at all, through a satisfaction survey after implementation.

Meanwhile, the market moves faster than the guidance. The Department for Education has issued advice on generative AI, and the Information Commissioner's Office has been clear about data protection expectations, but neither replaces local democratic oversight. Ofsted may ask about curriculum and safeguarding; it will not audit an algorithm. Multi-academy trusts centralise procurement, which can bring efficiency but also distance. The result is a gap between those who choose the system and those who live with it.

An AI scrutiny panel: a modest but real mechanism

Schools should establish an AI scrutiny panel, adapted to their size and phase. Its purpose is not to veto every purchase or to turn pupils into data protection officers. It is to introduce structured challenge at the points where decisions are made.

A workable model would include eight to twelve pupils drawn from across year groups, selected by election or sortition, with support from a governor, a senior staff lead and an independent facilitator where possible. The panel should meet at least termly and at three procurement gateways: before a pilot, before a contract is signed, and before renewal. Its remit should cover the summary data protection impact assessment, the vendor questionnaire, pilot feedback and any proposed use in assessment. It should report to the governing board or trust board, and its minutes should be published in an appropriate form.

The safeguards matter as much as the structure. Pupils should not access personal data about their peers. They should receive training in data protection, safeguarding and the limits of their role. They should be able to speak candidly without commercial confidentiality being used to silence them. And they should see what happens to their recommendations. A panel that meets once and is ignored is worse than no panel, because it teaches cynicism.

Questions pupils should put to vendors

A scrutiny panel needs a question set that is plain enough to use and rigorous enough to matter. Pupils should ask:

• What pupil data is collected, where is it stored, and for how long?
• Is pupil data used to train or improve the vendor's models, and can that be switched off contractually?
• Can pupils and parents see, correct or challenge the data held about them?
• How does the tool explain its outputs, and what evidence is there that it works fairly for pupils with SEND, EAL or disadvantaged backgrounds?
• Does the system make decisions or only recommendations? Who reviews those recommendations, and how?
• What happens if a school withdraws? Can all data be exported and deleted, with evidence?
• Are there audit logs, incident histories and clear breach notification arrangements?
• How is the tool integrated with the management information system, and what access does it have?
• If the tool is used in assessment, what is the appeals route and what weight does its judgement carry?

These are not exotic questions. They are the questions any responsible public body should ask before spending public money on a system that shapes children's experience.

Safeguards around data and assessment

Three safeguards should be non-negotiable. First, data minimisation: collect only what is necessary, retain it only as long as necessary, and do not allow function creep from teaching support into surveillance. Second, human review: no automated decision should determine a pupil's set, intervention, predicted grade or access to opportunity without a named professional able to explain and overturn it. Third, assessment transparency: if an AI tool informs assessment, pupils and parents should know how, and there should be a route to challenge error.

Schools should also publish a short annual AI register listing tools in use, their purpose and the safeguards applied. This is not burdensome transparency; it is basic accountability. Trusts should write data protection and audit clauses into contracts, and governors should ask for a summary of the data protection impact assessment rather than accepting a vendor's assurances.

Making it happen in 2026-27

No school needs to wait for perfect national guidance. A trust can pilot an AI scrutiny panel in one or two schools this academic year, link it to existing pupil voice structures and align its meetings with the procurement calendar. A governor can add AI to the agenda of the curriculum and standards committee. A headteacher can insist that pupils see a plain-English summary of any new tool before it is rolled out.

The deeper point is democratic. Pupils are told that their voices matter in school, in society and in public life. If the systems that increasingly shape their learning are chosen behind closed doors, that message is undermined. An AI scrutiny panel will not settle every question about technology and education. But it would begin to put the governance of school AI where it belongs: in the hands of the school community, including the pupils who must live with the consequences.