Sir Ian Bauckham, Chief Regulator at Ofqual, did not hedge. Speaking to the Confederation of School Trusts conference in Birmingham on 16 October, he asked what should be done about the misuse of AI in qualifications. “This is simple,” he said. “We cannot allow it.”
He went further than exam rules. Outsourcing student thinking to AI, he said, “does not develop the human mind”. It leads to less learning and less satisfaction, and is “an assault on the realisation of human potential”. Then the sharpest line in the speech: “It is less cognitive outsourcing – it is more like cognitive surrender.”
He was equally clear about what AI is for. Properly used, he said, it is helpful in easing the burden on teachers and improving the quality of curriculum and teaching resources. The problem is not AI in school. It is AI doing the pupil's thinking.
Coursework is the pressure point
Bauckham repeated a warning he has given before: extended writing coursework is vulnerable to cheating. He pointed to universities moving the other way from the AI-friendly assessment some advocate. Princeton has moved to invigilate all in-person exams; UCL Law School has made greater use of invigilated exams to make assessments more “AI‑proof”.
He also drew a lesson from smartphones. Over two decades, he said, young people were allowed undiscriminating access to phones, and we have seen the harm. We need to “not repeat the mistake with undiscriminating use of AI.”
A line on AI marking
The speech set out where Ofqual sees AI helping the qualifications system itself: writing exam questions, quality assurance and analysing performance. AI marking of complex work, where a system mimics human judgement of essays, artworks or performances, is different. Bauckham said it “opens up real ‘routes to harm’”: AI hallucinates, reflects biases and could extend a mark scheme to cases humans never agreed.
His conclusion was ethical as much as technical: “the dignity of the human person requires that consequential judgements made about a person's future need direct human oversight.”
This was a speech, not new regulation, and the published text is the original script. But it is a clear statement of where the regulator stands.
What schools can do with it
Check where your assessment relies on unsupervised writing. Look at coursework and non-exam assessment that depends on extended writing done away from the classroom. Where can drafting, planning or a final section happen under supervision? Where can a short conversation about the work confirm it is the pupil's own?
Teach the reason, not just the rule. Pupils who hear only “AI is banned in coursework” learn compliance. Pupils who understand why handing over their thinking means they learn less can make better decisions in every subject, including when no one is checking. Bauckham's own image helps: you don't get fit by watching a marathon on TV.
Use AI where it helps, openly. The same speech accepts that AI can ease teachers' workload and improve resources. Show pupils the difference between a teacher using AI to draft a quiz they then check, and a pupil using AI to write the answer they are supposed to work out.
Keep humans on consequential judgements. If your school is offered AI tools that grade extended work or make decisions about pupils, ask where a teacher reviews the outcome. That principle applies well beyond exams.
The phrase “cognitive surrender” is memorable because it names what is at stake. The job of AI literacy is to make sure pupils can tell the difference between getting help and giving up the thinking.
Source: Ofqual, Connecting curriculum, assessment and trust: qualifications for the next decade, speech by Sir Ian Bauckham, 16 October 2026. Published as the original script, which may differ from the delivered version. Applies to qualifications in England.