England’s Proposed EdTech and AI Safety Certification: What Suppliers Should Start Evidencing Now

Schools in England buy software the way many small businesses do: with a short questionnaire, a quick demo and a lot of trust. The Department for Education (DfE) now wants to replace some of that trust with an independent check. On November 10, 2026, it opened a consultation on a product certification scheme for EdTech and AI tools used in early years settings, schools and colleges. Under the plan, accredited assessors would test higher-risk products against published safety criteria, and certified products would go on a public register. The scheme is still only a proposal, and it covers England rather than the whole UK. But the draft criteria are detailed enough that suppliers can start building their evidence now.

What DfE is actually proposing

The consultation document explains the problem plainly. Schools and colleges are expected to judge technical safety risks with limited information from the market, and half of school leaders in DfE’s 2025 Technology in Schools Survey said safeguarding and data concerns held back their use of technology. Suppliers have the opposite problem: no recognised way to show that they meet safety expectations, so every sale repeats the same due diligence. DfE’s stated aims are to support safer decision-making, reduce the burden on settings and give suppliers “a clear and proportionate route” to show that their products meet defined safety standards.

The proposal follows the standard UK conformity-assessment model. DfE would own the scheme and set the criteria. Its preferred option is to license day-to-day delivery to one or more third parties. The United Kingdom Accreditation Service (UKAS) would accredit conformity assessment bodies against ISO/IEC 17065, the standard for bodies that certify products. Those bodies would assess products, charge suppliers a fee, issue certificates and handle complaints, suspensions and withdrawals. Certified products would carry a badge and appear on a public register showing what was assessed, who certified it and when the certificate expires.

Certification would be about safety, not quality. DfE says it will not judge whether a product improves learning, except where poor performance itself could put children at risk. That distinction matters for marketing teams. A safety badge isn’t evidence that a product works, and the consultation asks how to keep buyers from reading it that way. Our explainer on what a certification seal proves and what it does not makes the same point about GDPR seals.

Which products would be in scope

DfE proposes a risk-based scope, and a lot of everyday software is left out. General office tools, static content such as PDF textbooks, hardware such as tablets and whiteboards, and staff-facing tools that don’t carry out high-risk processing of children’s data are all excluded. AI agents or custom tools that teachers build for use only within their own school would also be exempt, though the setting would remain responsible for using them safely and lawfully. The four proposed categories are:

• Filtering products. These would be assessed against DfE’s existing filtering and monitoring standards. The criteria include implementing the Internet Watch Foundation URL list and the Counter-Terrorism Internet Referral Unit blocklist, making sure no administrator can disable illegal-content blocklists, keeping non-alterable admin logs and blocking VPN workarounds. Filtering data also couldn’t be reused to build a profile of the child. DfE is also asking whether filters should have to handle real-time, AI-generated content. Monitoring-only products are left out of this category for now.
• High-risk processing of children’s data. DfE borrows the ICO’s examples of processing likely to be high risk: large-scale profiling, special category or biometric data, data about vulnerable people, combining datasets, systematic monitoring, automated decisions with significant effects and innovative technology. Products likely to fall in scope include AI tutors, personalised learning systems, behaviour and safeguarding systems, SEND and wellbeing platforms, management information systems, and learning management systems that handle special category data.
• Student-facing generative AI. Certification would attach to the product children actually use, such as the app, chatbot or tutor, not the underlying model or API. A supplier that builds on a third-party model would still be responsible for certifying its own product.
• Student-facing personalised learning tools. These would be assessed against an adapted version of the generative AI standards, with a focus on pressure, gamification and engagement features that have no teaching purpose.

The criteria behind the badge

For products that carry out high-risk processing of children’s data, suppliers would need to show they have completed a data protection impact assessment, comply with UK data protection law and follow the relevant ICO codes. Those codes include the Age Appropriate Design Code, the planned ICO EdTech Code of Practice and, once it exists, a code on AI and automated decision-making. Suppliers would also need to show intellectual property compliance and adequate cyber security, with Cyber Essentials Plus given as an example. DfE is clear that certification would not replace the ICO, would not protect a supplier from enforcement and would not move enforcement to DfE. The Children’s Code still applies with or without a badge, as the ICO’s fine against Imgur’s parent company showed.

Student-facing generative AI tools would be tested against eight outcomes taken from DfE’s existing generative AI product safety standards: filtering, monitoring and reporting, design and testing, governance, cognitive development, emotional and social development, mental health, and manipulation. The three data criteria in those standards are left out of this strand, because every student-facing generative AI product would already be certified under the high-risk data category. The developmental criteria are the newest part. Suppliers would need to show they reduce the risk of cognitive deskilling and emotional dependence and respond safely in a mental health emergency, the same concerns that drive the debate over AI companions and children. Each certificate would record the product version, intended age range, configuration, known limitations and conditions of use.

Certification would also need to be renewed. DfE suggests full recertification every three years for filtering, data and personalised learning products, with annual checks in between. Generative AI tools might need recertifying every year. Any material change that affects safety or functionality would have to be reported to the assessor, who could then require a focused reassessment or a full recertification.

How suppliers can start building evidence

The consultation lists the assessment methods DfE expects to use: documentation review, technical review, benchmarking against set scenarios, safeguarding scenario testing and red teaming. Suppliers that work on these now will be better placed whatever the final scheme looks like.

1. Work out which category you fall into. Map each product and feature against the four categories. A product can fall into more than one. Write down why any feature is out of scope, because buyers and assessors will ask.
2. Update your DPIA so it reflects the real product. Cover profiling, any special category data, automated decisions and model providers. A structured DPIA template helps keep versions consistent across releases.
3. Check yourself against the Children’s Code. Look at default settings, data minimisation, nudge techniques, profiling switched off by default and geolocation. The ICO’s EdTech audit work found recurring gaps, including collecting more data than needed.
4. Build safeguarding test cases. For generative AI, keep a repeatable set of harmful prompts, jailbreak attempts, self-harm disclosures and emotional-dependence scenarios. Log each run and the result.
5. Write down which safeguards depend on the school. If some protections rely on platform settings or school configuration, DfE says the supplier would have to show how they work in practice.
6. Tighten security and logging. Get Cyber Essentials Plus or an equivalent, keep tamper-resistant admin logs and set up a supplier patch process. Breaches like the one behind the FTC’s action against Illuminate Education show why buyers care about this.
7. Set up change control now. Decide in advance what counts as a material change, such as a new model, new data use or new features for children, so that annual recertification doesn’t come as a surprise.

Early-stage products might get a lighter start. DfE floats a self-assessment or documentation review as a “readiness” route before full certification at scale, though it stresses that this would not count as certification. It also asks smaller suppliers how much they could pay each year, with options running from under £2,000 to over £10,000.

Why a voluntary badge may not stay optional for long

DfE has no power to make schools buy certified products or make suppliers seek certification, so the scheme would technically be voluntary. In practice it may not feel that way. DfE proposes adding an expectation to Keeping Children Safe in Education, with a matching expectation in the Early Years Foundation Stage framework, that settings get proportionate assurance for in-scope products. Certification would “ordinarily provide that assurance.” Settings could rely on other credible evidence, but they would have to explain why it is equivalent. Ofsted might look at these decisions as part of safeguarding, certification could become a condition for DfE procurement frameworks, and the consultation asks whether legislation should follow.

England isn’t the only place where regulators are writing rules for school technology. Spain’s data protection authorities, for example, published a decalogue for procuring digital educational platforms. What sets the English proposal apart is the use of independent, accredited assessment against criteria that cover child development as well as data protection.

Frequently asked questions

Does this apply to suppliers outside the UK?

Yes, if their products are used in English education settings and fall within scope. DfE specifically invites international EdTech suppliers to respond. The scheme follows the product, not the supplier’s location.

If we hold ISO 27001 or a GDPR seal, are we covered?

Not automatically. Existing certifications may count as evidence, and DfE says avoiding duplication is a priority. But the proposed criteria include child-specific safeguarding and developmental outcomes that general security or privacy certifications don’t test. The consultation asks suppliers which frameworks they already meet.

We use a third-party large language model. Who gets certified?

You do. DfE proposes certifying the student-facing product as configured, because the product supplier controls how the model is integrated and what protections are in place.

When could certification actually start?

DfE plans to publish its response in spring 2027. It then needs at least 18 months to set up the scheme before any product can be certified, followed by a transition period. Realistically, the first certificates are unlikely before late 2028.

Respond while the criteria can still change

Consultations like this one reward the suppliers who show up with evidence. DfE is asking very practical questions: what lead time you need, which costs you would absorb, how to test personalisation features, and at what scale full certification should start. Suppliers that answer with real data from their own DPIAs and test logs will help shape criteria they can actually meet. The full proposal and response form are on the GOV.UK consultation page, and the deadline is February 4, 2027.