Seldon report calls for federated pupil data model and "data lakes" to drive AI in schools

Educationalist Sir Anthony Seldon and venture capitalist Tim Bunting have called on the Department for Education to overhaul how pupil data is collected, held and shared, including a federated data model, institution-level "data lakes" and the exchange of education data with other countries within five years.

In a report entitled Britain's Greatest Education Opportunity Ever, published with strategy consultancy Public First, the authors described England's education system as having "no data in a data-driven society". Seldon, an historian and former head of Wellington College, said education "has always been Britain's Cinderella sector when it comes to technology, and with AI we are repeating the mistake at even greater speed and scale".

He added that uncertainty in schools, colleges and universities about how to respond "hasn't been helped remotely by a vacuum in leadership from government". The report is built on a Public First survey of 4,021 people aged 11 to 24 in England conducted between 18 and 27 August 2026.

The report's ten-point plan includes a rapid response unit on AI within the DfE, an AI strategy in every school approved by governors by September 2027, Ofsted inspection of those strategies and a "two-lane" assessment model combining terminal exams with continuous AI-supported assessment. Amongst these recommendations are:

• a federated data model in which anonymised data is held at setting level but moves with pupils as they progress through the system
• interoperable data fields, described as "a data structure that continues to learn precisely because it can be combined with others' data"
• institutional "data lakes" that "will inform decision-making and predictive behaviour of the institution and must be built"
• a target that "within five years, it should be plausible for the UK Government to exchange education data with high-performing countries, under the right permissions and safeguards"
• a recommendation that the education sector learn from the data practices of private data companies, with Palantir's pooling of health data cited as a model

The authors argue that the DfE's planned "data spine" is "hampered by arguments as to how much data can be collected, and the extent to which the DfE will be able to analyse it", and that "education certainly won't advance on the basis of the limited number of fields currently collected".

They set out the scope of what could be gathered: "a pupil's exact age, indicators of their social environment, performance in last formative assessment (down to question and sub-question level identifying hyper specific misunderstanding), performance of pupils around them, their keystrokes, even the room temperature when they sit an exam", amounting to "millions of data points about a single child: every essay written, every spelling test taken – even correlations with diet and sleep".

The report does not address the lawful basis for the expanded collection it proposes, the role of the ICO, consent or parental rights, or the data protection implications of international transfers of pupil data.

It also acknowledged that this approach has limits. "Data collection in education is bounded by sensitivities," it says. "It won't ever match the scale of data collected by retail giants or tech-first banking. Nor should it." It identifies "dehumanisation", in which "relentless drive for data displaces relationships and experiences at the heart of education", as one of three systemic risks of AI adoption, alongside cognitive offloading and the loss of academic integrity.

Data relating to children attracts specific protection under the UK GDPR, which requires that processing of children's personal data be given particular weight in any assessment of legitimate interests and that data protection impact assessments be carried out for large-scale processing of special category data or the systematic monitoring of individuals.

The ICO's Children's Code - which asserts that the best interests of the child must be a primary consideration in digital design - applies to online services likely to be accessed by children, and international transfers of personal data remain subject to Chapter V of the UK GDPR as amended by the Data (Use and Access) Act 2025. The report notes that the EU AI Act places education in its high-risk Annex III, covering admissions, assessment and monitoring, with the associated obligations postponed to December 2027.

The survey found that young people are already entrusting AI tools with personal information. Forty-one per cent of respondents said they would rather ask personal questions of an AI than of a person, and 40% had used AI for health advice. Forty-seven per cent said they trusted AI over people to check facts, against 21% who trusted people, and only 30% said their school had clear rules on AI use that they could explain. Eighty-seven per cent had used an AI tool, rising to 90% among 13 and 14-year-olds, with a 13-point gap between the highest social grade (92%) and the lowest (79%).

The full report is available at publicfirst.co.uk/seldon-bunting-report.