The use of Artificial Intelligence and EdTech in Education
Inquiry: House of Commons Education Committee, The use of Artificial Intelligence and EdTech in Education.
Submitter: Hleb Buziuk, independent public policy and governance researcher (personal capacity; no commercial interests).
Three key messages
● Current adoption intent is teacher-facing first, so near-term value is likely to be workload and administration, not proven learning gains. [1][3]
● Reported barriers increasingly include safeguarding, data and assurance, so governance must be usable by education buyers, not only written for suppliers. [2][3][10]
● Assessment integrity needs redesign and consistent reporting, not detection-only approaches. [5][6]
Key findings
Key recommendations (Who / What / When / How to check)
● Based on existing DfE safety standards and ICO guidance [2][10], DfE (with ICO, Ofsted and Ofqual) should publish a single AI and EdTech assurance pack for procurement and renewal, including standard DPIA prompts, supplier disclosures and incident routes, by Autumn 2026. Check: pack published and used in procurement processes.
● Based on current adoption patterns and barrier shifts [3], DfE should create an AI/EdTech Assurance Register for publicly funded settings, with declared use cases and minimum assurance artefacts, by Spring 2027. Check: register coverage, update cadence and sampling audit.
● Based on Ofqual’s regulatory position [6], Ofqual (with awarding organisations) should implement consistent AI-related malpractice categories and publish annual aggregate reporting by assessment mode by the 2026 to 2027 cycle. Check: time series published.
● Based on survey evidence of a training gap [3], DfE should fund minimum GenAI CPD for teachers, lecturers, DSLs and IT leads, with annual completion reporting from 2026 to 2027. Check: completion and confidence measures.
What is new vs official sources
● Computed, reproducible ratios that quantify adoption and delivery gaps. [3]
● A buyer-side accountability test across DfE, Ofsted, Ofqual and the ICO, with a concrete audit request. [2][5][6][10]
● A minimum data request pack to let the Committee test impact, safety and inequality claims (section 7).
Methods and limits: desk-based synthesis of published sources; no primary research (see section 3).
Signposts: key data requests in section 7; hearing questions in section 9; Evidence Ledger in Annex A (section 10).
I am an independent public policy and governance researcher. I submit in a personal capacity and have no commercial interest in AI or EdTech procurement. This evidence focuses on checkable published sources and on practical governance and data levers for Parliament.
● Desk-based review of published sources from DfE, Ofsted, Ofqual and the ICO, prioritising origin reporting and regulator guidance. [1][2][3][4][5][6][8][10]
● Headline numbers are anchored in the DfE Technology in Schools Survey where possible. [3]
● Derived metrics are computed transparently from published figures; inference claims are paired with a concrete test or data request (section 7).
● Coverage in the sources cited here is strongest for schools and some FE contexts; early years and higher education are treated as evidence gaps (section 7).
AI and EdTech are increasingly used for teaching, learning, assessment and safeguarding. This submission addresses opportunities and risks, infrastructure, impacts on teaching, assessment integrity, inequalities and children’s digital rights.
a) Headline paragraph
DfE has published product safety standards for generative AI, intended to shape supplier practice and procurement. [2] The ICO’s Children’s code and EdTech guidance clarifies that suppliers may be controllers where they determine purposes beyond a school’s instructions. [10] Ofsted states it does not evaluate AI tools directly, and focuses on outcomes and safeguarding practice. [5] In the DfE survey, safeguarding and data concerns as a barrier rose in primaries from 25% (2023, n=526) to 49% (2025, n=456). [3] The practical question is whether buyers have a single, usable route to evidence compliance at procurement and renewal.
b) Evidence-based observations
● Safeguarding and data concerns rose strongly in leader reporting since 2023. [3]
● Evidence gap: neither the DfE standards nor the ICO guidance provides a single buyer-ready checklist spanning procurement and renewal; the Committee could request one. [2][10]
c) Added value sentence
Combining [2], [5] and [10], the test is whether schools can evidence compliance without duplicating work.
d) Three points
● “Primary safeguarding and data barrier rose from 25% (n=526) to 49% (n=456).” [3]
● “Ofsted inspects outcomes, not AI tools.” [5]
● “EdTech roles depend on practice, not labels.” [10]
e) Counterpoint / trade-off
Over-complex assurance can raise costs and exclude smaller suppliers.
f) Options
● Option A: voluntary use of standards in procurement. [2]
● Option B: minimum mandatory assurance artefacts for procurement and renewal. [2][10]
Preferred: Option B, because it is checkable.
a) Headline paragraph
DfE frames teacher-facing use of generative AI as a near-term opportunity, while pupil-facing impacts are less evidenced. [1] Ofsted’s early adopter study describes use for administrative and planning tasks, often coordinated by an internal lead, alongside developing policies. [4] Survey evidence suggests a capability gap: 23% of primary teachers (n=797) and 35% of secondary teachers (n=414) report receiving GenAI training this academic year. [3] Leaders report GenAI training is more often planned than already offered. [3]
b) Evidence-based observations
● Teacher GenAI training reported this year is 23% (primary) and 35% (secondary). [3]
● Evidence gap: no published link between training completion and safer practice outcomes (data request in section 7).
c) Added value sentence
Triangulating [3] and [4], scaling depends on a minimum baseline, not local champions alone.
d) Three points
● “Teacher GenAI training is 23% primary and 35% secondary.” [3]
● “Early adopters often relied on an internal lead.” [4]
● “Teacher-facing benefits are more immediate than pupil-facing.” [1]
e) Counterpoint / trade-off
Mandating CPD without funding risks increasing workload.
f) Options
● Option A: guidance and exemplars only. [1][4]
● Option B: funded minimum GenAI CPD with annual reporting. [3]
Preferred: Option B, because coverage is low and measurable. [3]
a) Headline paragraph
Ofqual notes supervised examinations are less susceptible to GenAI misuse than non-exam assessment completed outside exam conditions, and states AI cannot be used as the sole marker of learner work. [6] Ofsted may consider how schools manage pupil AI use for homework and coursework where relevant. [5] The Curriculum and Assessment Review signals the need to consider AI in curriculum and assessment. [7] A key gap is comparable, published reporting on AI-related malpractice and mitigations by assessment mode.
b) Evidence-based observations
● Regulator expectations already constrain some AI uses in qualifications and marking. [6]
● Evidence gap: no routine, comparable annual dataset on AI-related malpractice by assessment mode (fields in section 7).
c) Added value sentence
Based on [5] and [6], the test is whether permitted use and disclosure rules are auditable.
d) Three points
● “Non-exam assessment is more exposed than supervised exams.” [6]
● “AI cannot be the sole marker of learner work.” [6]
● “Inspection may consider homework and coursework policies.” [5]
e) Counterpoint / trade-off
More supervised assessment can narrow curriculum and increase pressure.
f) Options
● Option A: strengthen guidance and rely on centre policies. [6]
● Option B: redesign assessment mix with authentication and disclosure rules. [6]
Preferred: Option B, to reduce ambiguity.
a) Headline paragraph
The DfE survey reports persistent barriers linked to pupil access at home, and reports stronger barriers in schools with higher proportions of pupils receiving free school meals. [3] The Connect the Classroom evaluation reports improved satisfaction with connectivity following upgrades, but continuing barriers, including hardware and software costs. [8] DfE digital and technology standards set expectations for infrastructure baselines, but they are not outcome measures. [9] The Committee may wish to focus on measurable access gaps and readiness thresholds before scaling learner-facing tools.
b) Evidence-based observations
● Home access barriers are widely reported and higher in high-FSM settings. [3]
● Evidence gap: this submission does not identify an origin dataset on access and outcomes for home educated children. Data request: include home education in the annual return (section 7).
c) Added value sentence
Combining [3], [8] and [9], the implementation test is whether minimum access thresholds are met.
d) Three points
● “Teachers report barriers on home device and connectivity access.” [3]
● “Connectivity upgrades help, but costs remain barriers.” [8]
● “Digital standards need readiness reporting to be useful.” [9]
e) Counterpoint / trade-off
Targeted access support may require sustained revenue funding.
f) Options
● Option A: continue infrastructure programmes without linked equity metrics. [8]
● Option B: publish disaggregated access and readiness metrics and target support. [3][9]
Preferred: Option B, because inequality risks are measurable. [3]
Quick wins (within 6 to 12 months)
1) Standard procurement and DPIA artefacts
● Who: DfE, with ICO input.
● What: based on existing standards and regulator expectations [2][10], publish templates and minimum supplier disclosures (data roles, retention, security, logging, incident routes) for tools used with under-18s.
● When: Autumn 2026.
● How to check: published templates and sample audit of adoption.
2) Funded minimum GenAI CPD
● Who: DfE and delivery partners.
● What: based on observed training gaps [3], minimum CPD for teachers and lecturers, plus DSL and IT lead modules on risk and incident response.
● When: 2026 to 2027 academic year; report annually.
● How to check: completion rates and confidence measures.
3) Assessment integrity package and reporting categories
● Who: Ofqual with awarding organisations, supported by DfE.
● What: based on Ofqual’s position [6], disclosure expectations, authentication options for non-exam assessment, and annual aggregate reporting of AI-related malpractice by assessment mode.
● When: by the 2026 to 2027 cycle.
● How to check: published categories and time series.
Structural actions (12 to 24 months)
4) AI and EdTech assurance pack and Assurance Register
● Who: DfE (with Ofsted, Ofqual and ICO).
● What: based on fragmentation across standards, inspection and data roles [2][5][6][10], publish a single assurance pack for procurement and renewal, and run a register of products used in publicly funded settings with declared use case and minimum assurance artefacts.
● When: pack by Autumn 2026; register by Spring 2027.
● How to check: coverage, update cadence and sampling audit.
Committee levers vs delivery actions
● Committee levers: request the minimum data pack (section 7); ask for ownership and timetable for the register; ask Ofqual for reporting plans; ask ICO what evidence buyers should require on data roles. [6][10]
● Delivery actions: DfE publishes templates, funds CPD and runs the register; regulators align guidance; settings maintain tool inventories and incident logs.
Metric 1: Planned AI investment skew (leaders)
● Inputs: 58% plan AI tools for teachers; 20% plan AI tools for pupils (leaders, n=795). [3]
● Computed value: ratio 2.9 (rounded to 1 decimal place); gap 38 percentage points.
● Recipe: Figure 8.12; compute 58/20 and 58 minus 20.
● Caveat: intention, not spend or impact.
Metric 2: GenAI training implementation gap (leaders)
● Inputs: primaries 21% offer vs 46% plan (leaders, n=456); secondaries 25% vs 53% plan (leaders, n=339). [3]
● Computed value: gaps 25 and 28 percentage points.
● Recipe: Chapter 7 GenAI training; compute plan minus offer.
● Caveat: does not measure quality or reach.
Field | Definition | Owner | Cadence | Enables |
Tool inventory | Tools, supplier, and use case | DfE via annual return; settings | Annual | Adoption and concentration |
Spend | AI and EdTech spend split by category | DfE via trusts and LAs | Annual | Value for money, equity |
Training completion | Completion by role | Settings; DfE aggregate | Annual | Capability tracking |
Incidents | Data, safeguarding, harmful output incidents | Settings; DfE aggregate | Termly | Risk monitoring |
Supplier artefacts | Data roles, retention, logging, security | Suppliers; DfE register | Annual update | Procurement assurance |
Malpractice | AI-related cases by assessment mode | Ofqual via awarding bodies | Annual | Assessment integrity |
Home access | Device and connectivity access at home | DfE via survey | Annual | Inequality monitoring |
Home education | Access and barriers for home educated children | DfE and LAs | Annual | Inclusion risks |
Use the fields in section 7 to test “impact” claims over time: spend and readiness; training completion; tool use by use case; incidents; malpractice; home access gaps. This can form a basic performance and value-for-money dashboard.
DfE
Ofsted
3) If Ofsted does not evaluate AI tools directly, what governance and safeguarding evidence will inspectors expect?
Ofqual
4) When will AI-related malpractice categories be implemented and published, and how will reporting be made comparable?
5) What assessment redesign options are most feasible without increasing inequity?
ICO
6) What common EdTech compliance failures do you see, and what procurement disclosures would prevent them?
● Inequality: home access gaps are widely reported. [3] We infer a risk of widening attainment differences.
● Safeguarding and privacy: unclear data roles and retention practices. [10]
● Assessment integrity: Ofqual notes non-exam assessment is more exposed than supervised exams. [6] We infer monitoring is harder without comparable reporting.
● Workload: we infer compliance burden could shift to DSLs and IT leads.
● Cyber security: we infer expanded attack surface via new tools and integrations.
● Vendor lock-in: we infer fragmented procurement can increase switching costs.
● Publish the minimum data pack (section 7) to enable evaluation.
[1] Department for Education. Generative artificial intelligence (AI) in education. 12 August 2025. https://www.gov.uk/government/publications/generative-artificial-intelligence-in-education
[2] Department for Education. Generative AI: product safety standards. 19 January 2026. https://www.gov.uk/government/publications/generative-ai-product-safety-standards
[3] Department for Education. Technology in Schools Survey 2024 to 2025: research report. November 2025. https://assets.publishing.service.gov.uk/media/692834a6ce50d215cae9610e/Technology_in_schools_survey_2024_to_2025_research_report.pdf
[4] Ofsted. Insights from early adopters of artificial intelligence in schools and further education colleges. 27 June 2025. https://www.gov.uk/government/publications/ai-in-schools-and-further-education-findings-from-early-adopters/the-biggest-risk-is-doing-nothing-insights-from-early-adopters-of-artificial-intelligence-in-schools-and-further-education-colleges
[5] Ofsted. How Ofsted looks at AI during inspection and regulation. 27 June 2025. https://www.gov.uk/government/publications/ofsteds-approach-to-ai/how-ofsted-looks-at-ai-during-inspection-and-regulation
[6] Ofqual. Ofqual’s approach to regulating the use of artificial intelligence in the qualifications sector. 24 April 2024. https://www.gov.uk/government/publications/ofquals-approach-to-regulating-the-use-of-artificial-intelligence-in-the-qualifications-sector
[7] Department for Education. Curriculum and Assessment Review: final report. 5 November 2025. https://www.gov.uk/government/publications/curriculum-and-assessment-review-final-report
[8] Department for Education. Connect the Classroom: evaluation. 11 February 2026. https://assets.publishing.service.gov.uk/media/698c75e82f683cc788c28774/Connect_the_classroom_evaluation.pdf
[9] Department for Education. Meeting digital and technology standards in schools and colleges. 23 March 2022. https://www.gov.uk/guidance/meeting-digital-and-technology-standards-in-schools-and-colleges
[10] Information Commissioner’s Office. The Children’s code and education technologies (edtech). 30 May 2023. https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/childrens-information/childrens-code-guidance-and-resources/the-children-s-code-and-education-technologies-edtech/
● Annex A: Evidence Ledger (three headline findings).
Claim | Claim type | Source A | Source B | Source B role | Status | Confidence | What would change my mind | Verification owner |
Leaders plan more AI investment for teachers than pupils (58% vs 20%, n=795), ratio 2.9. | SYNTHESIS (derived) | [3] Figure 8.12 | [1] | Context only | ORIGIN-ANCHORED | High | Spend data shows no skew | DfE |
GenAI training is planned more than offered (21% vs 46% primary; 25% vs 53% secondary). | SYNTHESIS (derived) | [3] Chapter 7 | [4] | Independent corroboration (qual) | ORIGIN-ANCHORED | Medium-High | Admin completion data shows high coverage | DfE / settings |
Safeguarding and data barrier rose in primaries from 25% (2023, n=526) to 49% (2025, n=456). We infer there is no single auditable buyer assurance route across standards, inspection and data roles. Test: audit procurement packs and DPIAs against [2] and [10]. | INFERENCE (with test) | [3] Table 8.1 | [2][5][6][10] | Independent corroboration (roles differ) | INPUTS-ANCHORED | Medium | Audit shows consistent assurance route | DfE / regulators |
May 2026
Submission 1 of 2 (Audit, metrics, value for money & accountability)