Research Question

Where do current DfE AI guidance documents contradict or leave ambiguity around acceptable use in summative assessment, and what risks does this create for schools?

Methodology

This case study employed systematic policy document analysis to identify gaps and ambiguities in UK government guidance on AI use in summative assessment. The research examined Department for Education publications on generative AI in education (2025), Joint Council for Qualifications instructions for conducting coursework (2024–25), exam board non-exam assessment guidance from AQA, Pearson and OCR (2024–25), Ofqual policy communications on AI detection tools (2024), and Ofsted Education Inspection Framework applications to assessment integrity (2024).

Documents were analysed for explicit policy positions on AI use in assessment contexts, definitional clarity around “legitimate support” versus “unauthorised assistance”, implementation guidance for schools, and consistency across regulatory bodies. Cross-referencing identified contradictions, silences and interpretive challenges schools must navigate without institutional support.

Executive Summary

Key findings: DfE guidance promotes AI as an educational tool whilst JCQ defines unauthorised use as malpractice, with no threshold clarifying the boundary between legitimate support and academic misconduct. Policy delegates interpretation to teacher judgement whilst examination regulations demand standardised, auditable controls, creating incompatible expectations. Exam boards provide varied emphases (monitoring, declarations, supervision) for identical qualifications, leaving schools to resolve interpretive differences. Detection tools are acknowledged as unreliable yet no alternative frameworks exist, creating impossible evidential standards for malpractice reporting.

Critical risks: Schools face exam board sanctions for both over-reporting (insufficient evidence) and under-reporting (missed violations), parental appeals citing DfE’s pro-innovation stance, Ofsted scrutiny of assessment integrity without AI-specific guidance, and institutional inconsistency across departments implementing contradictory policy signals.

The Compliance Minefield

It’s June 2026. A Year 11 student submits GCSE coursework. Your English teacher flags potential AI use. The analytical depth seems beyond the student’s classwork, but there’s no smoking gun. You check DfE guidance. One section suggests AI is a legitimate learning tool. Another warns about academic dishonesty. A third defers to exam boards, who each interpret “permissible assistance” differently. Your decision could trigger an exam board investigation, parental complaint, or Ofsted question about your assessment integrity.

This isn’t hypothetical. It’s the compliance minefield UK schools navigate daily because current DfE guidance on AI in summative assessment contains significant gaps and unresolved ambiguities.

Think about what you’re being asked to do: identify AI use in student work whilst being told detection tools don’t work, apply professional judgement whilst exam boards demand documented evidence, encourage AI as a learning tool whilst treating it as potential malpractice, and maintain institutional consistency whilst navigating contradictory policy signals.

You’re operating in regulatory uncertainty requiring sophisticated risk management, not straightforward compliance.

What We Found: Four Policy Gaps Creating Impossible Positions

Gap 1: Educational Tool Versus Academic Misconduct

The DfE’s 2025 guidance on generative AI in education positions AI as beneficial for workload reduction and personalised learning, actively encouraging schools to embrace its opportunities whilst developing context-specific policies aligned with safeguarding requirements (Department for Education, 2025). This pro-innovation stance pervades departmental communications, from leadership toolkits to ministerial blog posts.

Yet the Joint Council for Qualifications’ Instructions for Conducting Coursework 2024–25 unequivocally states that unauthorised AI use in controlled assessments constitutes malpractice subject to sanction (Joint Council for Qualifications, 2024).

The critical gap: neither document defines where “legitimate learning support” ends and “unauthorised assistance” begins.

A Year 10 student using AI to generate essay plans during homework? The DfE framework encourages this exploration. The same student applying identical techniques in coursework? Potentially reportable malpractice under JCQ regulations.

Schools must define this boundary without policy guidance, knowing that incorrect interpretation risks either unreported malpractice or wrongful sanctions. You’re expected to draw a line the government won’t define, then defend your position when exam boards, parents, or inspectors challenge it.

Gap 2: Teacher Autonomy Versus Evidential Standards

DfE guidance emphasises teacher professional autonomy in determining appropriate AI use within subject contexts, a decentralised approach respecting pedagogical diversity (Department for Education, 2025). Computing teachers might legitimately teach prompt engineering; English teachers might use AI for drafting practice. This flexibility recognises that AI’s educational value varies across disciplines.

However, JCQ regulations demand documented evidence trails for all controlled assessment decisions (Joint Council for Qualifications, 2024). When Ofsted inspectors question assessment integrity, answering “We trust individual teacher judgement” without school-wide protocols documented against the Education Inspection Framework creates compliance vulnerability.

The DfE promotes flexibility. Exam regulations require standardisation. Schools operate in the tension between these incompatible expectations with no reconciliation mechanism.

Here’s the impossible position: you’re told to trust professional judgement whilst simultaneously required to produce audit trails demonstrating standardised decision-making. Professional judgement becomes professionally defensible only when documented through processes that contradict the autonomy you’re supposedly exercising.

Gap 3: The Exam Board Interpretation Problem

The DfE explicitly delegates specific assessment rules to exam boards, defensible given that AQA, Pearson and OCR serve different educational contexts. In practice, this creates interpretive challenges for schools navigating identical qualifications.

Exam board non-exam assessment guidance reveals variation in emphasis: AQA stresses monitoring processes within existing frameworks, Pearson explicitly requires student declarations of AI use, whilst OCR mandates teacher supervision without defining what constitutes adequate oversight parameters (AQA, 2024; Pearson, 2024; OCR, 2024).

Multi-academy trusts operating across regions must interpret these emphases differently for the same GCSE specifications. A student taking AQA English Literature in one school faces different AI monitoring than their cousin taking the identical specification with Pearson in another trust school.

Schools must manually audit each qualification’s AI-specific guidance, then operationalise variations the DfE won’t harmonise. This individualises policy risk that should be resolved at system level. You’re doing compliance work that belongs with the regulator, without the regulator’s authority to make definitive interpretations.

Gap 4: Detection Rhetoric Without Detection Standards

DfE guidance acknowledges teachers’ role in maintaining summative assessment integrity, implicitly expecting AI-use identification as professional duty (Department for Education, 2025). Simultaneously, the department notes that AI detection tools produce unreliable results unsuitable for high-stakes decisions.

Ofqual communications reinforce this position, noting that detection software generates false positives rendering it unfit for malpractice evidence, yet offering no alternative framework for schools to use instead (Ofqual, 2024). Meanwhile, JCQ standards require proof beyond teacher suspicion before reporting potential violations (Joint Council for Qualifications, 2024).

Schools face an impossible standard: you’re expected to identify AI use but denied reliable tools, whilst teacher “professional judgement” doesn’t constitute sufficient evidence for formal reports.

Over-reporting based on suspicion invites exam board challenges for insufficient evidence. Under-reporting risks centre malpractice investigations if violations are later confirmed. There’s no compliant middle ground. You’re damned for acting on suspicion, damned for not acting, with no guidance on how to navigate between those positions.

The Risk Landscape: Four Material Threats

Exam Board Sanctions

Schools reporting suspected malpractice based on DfE’s emphasis on professional judgement may face exam board rejection if evidence doesn’t meet JCQ’s evidential standards. The inverse risk is equally severe: failing to report later-confirmed AI use triggers centre malpractice investigations, potentially including qualification withdrawal and future assessment sanctions (Joint Council for Qualifications, 2024).

A teacher’s conviction that work appears AI-generated doesn’t satisfy reporting thresholds. Waiting for definitive proof means violations go unreported. Schools navigate this without policy cover for either decision.

Parental Appeals

Parents increasingly cite DfE guidance when challenging school AI sanctions, arguing students engaged with technology “as a learning tool” consistent with departmental encouragement (Department for Education, 2025). Schools defending disciplinary decisions must reconcile their position with policy documents explicitly promoting AI exploration.

This creates legally vulnerable positions in complaints procedures. Your disciplinary policy prohibits what DfE guidance encourages. Parents recognise the contradiction; governors and appeals panels will too. You’re enforcing rules contradicted by government guidance, then expected to defend that position without policy support.

Ofsted Scrutiny

Inspectors now routinely question AI assessment safeguards during curriculum deep dives. The Education Inspection Framework’s quality assurance expectations require documented institutional responses to emerging assessment risks (Ofsted, 2024). Whilst Ofsted provides no AI-specific guidance, inspectors apply general integrity principles to emerging technologies, expecting schools to demonstrate proactive risk management.

Schools without AI policies appear negligent. Schools with policies often can’t demonstrate how they’ve reconciled DfE’s pro-innovation stance with JCQ’s malpractice definitions, inviting extended dialogue about leadership effectiveness and curriculum integrity. Either position weakens inspection outcomes.

Institutional Inconsistency

Without clear frameworks, individual teachers apply divergent standards. One department permits AI research tools, another prohibits any AI contact during assessed work. Students experience this as inequitable, parents as unprofessional.

When Ofsted reviews assessment consistency across subjects, these variations signal inadequate quality assurance. Yet teachers are implementing contradictory guidance from credible authorities. The DfE encourages AI exploration whilst JCQ defines it as potential malpractice. Your staff aren’t failing to follow guidance. They’re following different guidance pointing in opposite directions.

Recommendations: Managing Policy Uncertainty

Regulatory clarity isn’t imminent. Education policy typically lags technological change by three to five years; generative AI entered classrooms eighteen months ago. Schools will navigate ambiguity throughout multiple examination cycles, requiring strategic risk management rather than policy compliance.

Document your institutional synthesis. Create an AI assessment policy that explicitly states how your school has synthesised DfE and JCQ positions for your context. Reference the policy documents you’ve analysed and the reasoning behind your decisions. Include your definition of “legitimate AI support” versus “unauthorised assistance” for different assessment types, subject-specific variations justified by pedagogical requirements, the evidential standard triggering malpractice reports, and how you’ve balanced DfE’s innovation encouragement with JCQ’s integrity requirements.

This won’t eliminate uncertainty, but documented analytical processes demonstrate institutional due diligence during inspections, exam board inquiries or complaints procedures. Schools with reasoned frameworks defend their decisions; schools without them appear reactive.

Map qualification-specific requirements. Audit every GCSE and A-level specification for AI-specific regulations. Where exam boards provide different emphases for comparable qualifications, escalate collectively through multi-academy trust structures or local authority networks. Coordinated requests from multiple centres might accelerate board-level policy development. At minimum, collective escalation documents that you’ve identified the ambiguity and sought resolution, valuable during inspections or appeals.

Shift from detection to declaration. Since reliable AI detection remains technologically unfeasible (Ofqual, 2024), build assessment processes around student disclosure rather than teacher identification. Require declaration statements for all controlled assessments: “I confirm that any AI tools used in producing this work are listed below with explanation of how they supported my thinking. I understand that undeclared AI use may constitute malpractice.”

This creates accountability without depending on unreliable detection technology. It shifts responsibility appropriately: students declare, teachers verify declarations against visible work characteristics, malpractice investigations focus on undisclosed use rather than proving AI involvement.

Develop staff capacity for ambiguity. Teachers require professional development on navigating policy uncertainty, not waiting for definitive guidance that isn’t coming soon. Scenario-based training develops judgement: “A Year 9 student asks if AI can help structure their history essay. What questions clarify whether this enhances or replaces their analytical thinking? How do you guide them towards legitimate support?”

Professional judgement improves through deliberate practice with realistic dilemmas, not passive consumption of policy documents.

Meta Pedagogy Support

We help schools navigate contradictory AI assessment guidance by building defensible frameworks that protect both educational innovation and institutional compliance.

Policy gap analysis: Audit your current AI assessment policies against DfE guidance, JCQ regulations, exam board variations and Ofsted framework expectations. Identify where your institutional position is vulnerable to challenge from exam boards, parents, or inspectors. Map contradictions you’re expected to resolve without regulatory support.

Institutional synthesis support: Develop AI assessment policies that explicitly reconcile DfE and JCQ positions for your context. Document the reasoning behind your decisions, creating defensible frameworks that withstand inspection scrutiny whilst protecting educational innovation. Define “legitimate support” versus “unauthorised assistance” with subject-specific variations justified by pedagogical requirements.

Qualification-specific mapping: Audit every GCSE and A-level specification for AI-specific regulations. Create implementation guides showing how AQA’s monitoring emphasis differs from Pearson’s declaration requirements and OCR’s supervision mandates. Build processes ensuring consistent application across qualifications despite board-level variations.

Declaration-based assessment frameworks: Design assessment processes around student disclosure rather than unreliable teacher identification. Develop declaration templates, verification protocols, and malpractice investigation procedures focusing on undisclosed use rather than proving AI involvement.

Staff CPD on policy uncertainty: Scenario-based professional development building teacher capacity to navigate ambiguity through realistic dilemmas requiring professional judgement. Develop decision-making frameworks that work despite contradictory guidance.

Our honest approach: These policy gaps won’t resolve quickly. We don’t claim perfect solutions. We help you build robust frameworks despite policy uncertainty, not because we’ve identified definitive answers, but because documented analytical processes defend imperfect decisions when challenged. Institutional advantage belongs to schools that conduct rigorous work reconciling contradictions, not schools waiting for clarity that isn’t coming.

Need support auditing your AI assessment policies against current guidance? We’ve mapped the compliance landscape and can help you build frameworks that protect educational innovation whilst managing institutional risk during regulatory uncertainty.

Conclusions

Current DfE guidance on AI in summative assessment contains significant gaps and unresolved ambiguities creating compliance risks for schools. The DfE promotes AI as educational tool whilst JCQ defines unauthorised use as malpractice, with no definitional threshold between legitimate support and misconduct. Policy delegates interpretation to teacher judgement whilst examination regulations demand standardised controls. Exam boards provide varied emphases for identical qualifications. Detection tools are acknowledged as unreliable without alternative frameworks.

Schools face material risks: exam board sanctions for both over-reporting and under-reporting, parental appeals citing contradictory guidance, Ofsted scrutiny without AI-specific frameworks, and institutional inconsistency implementing contradictory signals.

Regulatory clarity isn’t imminent. Schools must navigate ambiguity through documented institutional synthesis, qualification-specific mapping, declaration-based accountability, and staff capacity development for policy uncertainty. Institutional advantage belongs to schools building defensible frameworks despite policy gaps.

References

AQA (2024) Non-exam assessment guidance 2024–25. Manchester: AQA.

Department for Education (2025) Generative artificial intelligence (AI) in education. Available at: https://www.gov.uk/government/publications/generative-ai-in-education

Joint Council for Qualifications (2024) Instructions for conducting coursework: 2024–25. London: Joint Council for Qualifications.

OCR (2024) Non-exam assessment guidance 2024–25. Cambridge: OCR.

Ofqual (2024) Policy communications on AI detection tools. Coventry: Office of Qualifications and Examinations Regulation.

Ofsted (2024) Education Inspection Framework. Manchester: Ofsted.

Pearson (2024) Non-exam assessment guidance 2024–25. London: Pearson.

Research case study completed: December 2025 | Word count: 2,000