At the center of the storm is the Joint Council for Qualifications (JCQ), the body overseeing major UK exam boards, which published an updated revision of its assessment integrity guidance on November 20, 2026. While the tabloid headlines conjure images of overworked teachers pasting GCSE essays straight into consumer chatbots and letting an algorithm spit out final grades, the actual regulatory text—and the technical reality behind it—is considerably more nuanced, and raises urgent operational headaches for school IT administrators and data protection officers.
What the JCQ Guidance Actually Says vs. the Headlines
To understand what changed, it helps to separate the rhetoric from the rulebook. In its revised document, AI Use in Assessments: Your role in protecting the integrity of qualifications (effective from September 2026, Revision one: November 2026), JCQ restructured its marking section to clarify two completely different scenarios:
- Marking student work that includes acknowledged AI use: Assessors must ensure candidates are not rewarded if AI tools prevented them from independently demonstrating the required knowledge, skills, or understanding. Whenever acknowledged AI use affects a mark or grade, teachers must keep clear records to support student feedback, internal appeals, or moderation.
- Using AI tools to support the teacher's marking: Here is the clause triggering the uproar. JCQ states that exam centres—schools and colleges—"may determine, after careful consideration of any data privacy concerns, whether it is appropriate for their teachers and assessors to use AI tools to help mark student work".
Crucially, JCQ attaches strict strings to that discretion. Where a centre allows AI support, an AI tool cannot be the sole marker. A human assessor must still review the student's work in its entirety, determine the mark using their own professional human judgment regardless of what the AI suggests, and remain personally accountable for the final grade.
In other words, JCQ has not mandated AI marking, has not endorsed any specific chatbot, and has not authorized automated grading. Rather, it has formally acknowledged that schools can allow AI-assisted marking workflows—provided they clear a data-privacy hurdle and keep a human firmly in the driver's seat.
| Policy Area | Rules for Students (Coursework / NEA) | Rules for Centres & Teachers (Marking) |
|---|---|---|
| Permitted Use | Allowed only if acknowledged, referenced, and compliant with subject rules; cannot earn credit for non-independent work. | Centre may decide whether AI tools can support marking after vetting data privacy. |
| Sole / Autonomous Role | Submitting AI-generated work as one's own is malpractice/plagiarism. | AI cannot be the sole or primary marker; human must review full work. |
| Audit & Records | Must retain non-editable screenshots of prompts, outputs, and explanation of use. | Must retain records where student AI use affects grades; teacher owns final mark. |
| Device & Network Controls | Online AI tools must be restricted on centre devices used for exams. | Subject to centre data-protection and safeguarding governance. |
Why Critics Call It a "Red Line" and a Double Standard
Even with those human-in-the-loop caveats, educationalists argue the guidance creates both an ethical contradiction and a reliability hazard.
Speaking to the Daily Mail, Dr Will Van Reyk—Director of Innovation at St Paul's Girls' School in London, which bans AI marking, and co-founder of the AI literacy platform Pedagogue—called the guidance "preposterous" and "nothing short of scandalous." His objection centers on three practical flaws:
- No approved toolset or technical standard: Dr Van Reyk warned that the JCQ wording leaves schools free to feed student coursework into general-purpose chatbots "like ChatGPT or Microsoft Copilot"—consumer-facing large language models (LLMs) never designed or validated as exam markers. (It is worth noting that JCQ's text neither names nor approves Microsoft Copilot or ChatGPT, leaving tool vetting entirely to individual centres.)
- Zero empirical backing for mainstream LLMs: "There is absolutely no evidence that a mainstream AI chatbot can be used for marking," Dr Van Reyk told the paper, warning that the practice amounts to "putting the kids' marks into a lottery."
- The student-teacher trust deficit: While Ofqual noted in a November 18, 2026 blog post that more than 1,100 students lost an entire GCSE or A level in 2025 over coursework malpractice, parents and students are not automatically told if a teacher ran their coursework through an AI tool. Dr Van Reyk argued that if teachers use AI to mark while telling pupils not to use AI to write, coursework becomes "totally invalid" at both ends.
Dr Van Reyk also accused Education Secretary Lucy Powell of contradicting the official guidance. In oral evidence before the Commons Education Select Committee on November 4, 2026, Powell told MPs: "Ofqual is really clear, as am I, that marking and the awarding of qualifications really should be done by the human eye, because otherwise you can exacerbate prejudice and other issues and in awarding them".
In response to the Daily Mail report, spokespeople for JCQ, Ofqual, and the Department for Education (DfE) all stressed alignment around the same guardrail: AI may help reduce teacher workload if used safely, but it must never serve as the sole or primary mechanism for determining a student's mark or replace a teacher's professional judgment.
Ofqual's Own Research Backs the Technical Skepticism
Here is the irony buried in the official paperwork: when it comes to the technical limitations of AI marking, England's exam regulator largely agrees with the critics.
In a detailed working paper published in January 2026 (Principles of AI use in marking) and an accompanying blog post on January 14, 2026, Ofqual laid out a sobering assessment of LLMs and deep neural networks in high-stakes grading:
- No general case for validity: Ofqual concluded that current evidence "does not support an overall or general case for the validity of AI marking in high stakes qualifications," warning that performance in one subject or item type does not reliably transfer to another.
- The "black box" appeals problem: Public confidence in qualifications depends on a student's right to challenge a mark and see why it was awarded. Ofqual noted that modern LLMs lack true semantic understanding and often cannot produce reliable, explainable justifications for why a specific mark was given.
- Algorithmic bias and drift: AI models can amplify demographic biases present in training data or freeze marking standards to a historical snapshot, failing to reflect how curriculum expectations evolve.
- Where Ofqual sees promise: Rather than autonomous grading, Ofqual's July 2026 regulatory framework highlights AI's potential in lower-stakes classroom feedback, marker training, and quality-assurance checks on human examiners.
That technical reality exposes a subtle psychological trap for any teacher using an AI assistant as a "second pair of eyes": anchoring bias. Even if a human teacher reads the entire essay as JCQ requires, seeing an AI chatbot's suggested score or rubric critique beforehand can subtly skew human judgment—especially under end-of-term marking fatigue.
The IT Admin and Data-Privacy Minefield for Schools
For IT administrators, Data Protection Officers (DPOs), and Microsoft 365 managers working in UK schools and colleges, the sharpest sting in JCQ's November 2026 revision is five words long: "after careful consideration of any data privacy concerns".
By delegating AI marking permissions to individual exam centres without specifying a list of compliant software, JCQ has effectively dropped a compliance and cybersecurity governance challenge onto school IT teams. Consider what happens the moment a teacher pastes student coursework into an unvetted web chatbot:
- UK GDPR and pupil PII exposure: Coursework frequently contains student names, candidate numbers, school identifiers, and sometimes deeply personal reflections or special educational needs context. Under DfE data-protection guidance for generative AI (updated October 2026), schools in England are explicitly instructed not to enter pupils' personal data into generative AI tools without consulting their DPO or IT lead, reviewing data-retention and model-training terms, and updating privacy notices.
- Consumer chatbots vs. enterprise-managed tenants: There is a vast technical difference between an unmanaged personal ChatGPT or consumer Copilot session—where prompts may be logged or retained under consumer terms—and a school-managed enterprise AI deployment (such as an organization-licensed Microsoft 365 Copilot tenant or sandboxed DfE-approved tool) governed by data-processing agreements and zero-training guarantees. Without explicit IT policy, teachers rarely distinguish between the two.
- Endpoint and exam-device lockouts: JCQ's guidance also reminds centres that online AI tools must be restricted on centre devices used during examinations and controlled assessments, requiring active web filtering, browser policies, and endpoint management.
- Audit trails and version history: To verify that student work is authentic—and to document how teacher marking decisions were reached—schools increasingly rely on draft logs, document version histories (such as those tracked natively in Word/OneDrive or Google Workspace), and secure non-editable prompt records.
Although JCQ's November 2026 update expanded its recommended staff training guidance to include practical experience with AI tools and understanding their limitations, recommendations are not binding technical standards. Until regulators or the DfE establish clearer technical baselines for classroom assessment tools, school leaders and IT admins are left drawing the line themselves: either enforce a blanket ban on AI-assisted coursework marking, as St Paul's Girls' School has reportedly done, or build strict, DPO-approved technical guardrails before a single essay ever hits a prompt box.
References
- Teachers are 'quietly' given official permission to use AI chatbots when marking exam coursework in 'scandalous' move, critics say - Daily Mail Daily Mail · 2026-11-21T21:10:34+00:00
- Using AI in marking: why technical capability, fairness, and transparency all matter – The Ofqual blog ofqual.blog.gov.uk
- AI and coursework: how we can tackle this together – The Ofqual blog ofqual.blog.gov.uk