Ofqual's chief regulator argues against letting AI mark essays and other complex exam answers

Ofqual's chief regulator has set out why he thinks AI should not be left to mark essays, artworks and performances. Sir Ian Bauckham, Chief Regulator at Ofqual, made the case in a speech to the Confederation of School Trusts annual conference at the ICC in Birmingham, listed as delivered on Friday 16 October 2026 and published on gov.uk the same afternoon.

gov.uk labels the text "Original script, may differ from delivered version", so the quotes below are from the script as written, not a transcript of what was said on the day.

Where he says AI does have a place

The speech is not a blanket rejection of AI in exams. Bauckham said: "There are certainly areas where AI can improve efficiency, cut costs and increase accuracy." He listed examples: "Helping to write exam questions, quality assurance, analysing performance, and so on."

He also drew a line between two things that are often lumped together: "To be clear, I am talking about AI marking here, not machine marking of questions with a very limited range of easily defined right answers."

On teaching, he said: "While AI – properly used – is helpful to ease the burden on teachers, and to improve the quality of curriculum and teaching resources, outsourcing student thinking to AI does not develop the human mind."

Near the end he summed up: "Of course AI has a place. But we must remain in charge."

His case against AI marking

The target is what he called AI marking: "AI marking, by contrast, is where we ask AI to mimic human judgement of complex responses. Essays, artworks, performances."

He gave several reasons:

• Errors and bias. "AI hallucinates, and it reflects biases. All current research supports this."
• The mark scheme drifting. He said "generative AI, by definition, has the potential capacity to ‘improve’ itself. To act autonomously." He added that "it could extend a human-agreed mark scheme to deal with cases not covered by that mark scheme. And at that point the expert human is no longer in the loop, even indirectly."
• Human oversight. "In my view the dignity of the human person requires that consequential judgements made about a person’s future need direct human oversight."
• Challenging a mark. "Moreover, using generative AI in marking in this way also makes the basic right to contestability very difficult, if not impossible, to exercise." Of human examiners, he said: "But at least you can challenge human errors. How do you challenge a judgement made by a self-evolving black box that no-one really understands?"

He framed the issue as more than a question of performance: "So in short, I think the AI question goes beyond efficiency, accuracy and cost savings, important and desirable as those may be. This is fundamentally an ethical question."

He also said: "We must legislate and regulate to be clear about what AI can be allowed to do, to bring about genuine benefits for humans, and what it absolutely cannot." The script does not say what that legislation or regulation would contain.

Cheating and coursework

On students using AI to cheat, he said: "This is simple. We cannot allow it." He said: "Extended writing coursework is vulnerable to cheating, as I have said many times before." He pointed to universities making more use of invigilated exams, naming Princeton University and UCL Law School.

He argued that using AI to skip the work harms learning: "It is less cognitive outsourcing – it is more like cognitive surrender." The script says research shows students who use AI to shortcut learning "achieve less well", but does not name a study. (In August we covered one such study of about 27,000 Chinese secondary students. It was a single study in one country's exam system and was not a randomised trial.)

The speech also covered V Levels and on-screen exams. On the latter he said: "Expect to hear more in due course."

How this fits Ofqual's existing rules

The speech matches a line Ofqual has already drawn. In its policy paper on regulating AI, updated on 16 July 2026, Ofqual says: "We have made clear to awarding organisations that AI may not be used as a sole marker in any assessments that are part of regulated qualifications."

The same paper says Ofqual applies "our existing, outcomes‑based regulatory framework to the use of AI, rather than introducing new AI‑specific rules", and that "we recognise the potential for AI to support marking of lower‑stakes and classroom-based assessment." In January 2026 Ofqual also published a working paper, Principles of AI use in marking, which it describes as being meant "to support thinking and foster constructive discussion about the potential applications of AI in marking".

Why it matters

On our reading, the speech does not change Ofqual's rules: its published ban on AI as a sole marker already stands. What it adds is the chief regulator's own reasoning for keeping humans in charge of judging complex work, including that consequential judgements about a person's future need direct human oversight and that a student must be able to challenge a mark. It also sets aside machine marking of short, fixed-answer questions from that objection, and names question-writing and quality assurance among areas where he says AI can help. The speech says further legislation and regulation are needed but does not set out what they would be.