Picture a Norwegian classroom in late August. The new school year is just beginning but this year, something is missing. The laptops and tablets are still there, but for students in Years K-7 the AI function has been abruptly switched off and for students in Years 8-10, AI use is only allowed under strict supervision. No chatbot to draft their stories, no generative assistant to solve their maths problems. Prime Minister Jonas Gahr Støre, announcing the policy this June, put it in the plainest terms a parent could understand: “The most important thing in school is that our children learn to read, write and do mathematics.”
It is a striking image, a government legislating not for more technology in the classroom but for less of it. It is also, for a British reader, a slightly uncomfortable one. Britain has spent the last two years debating AI in schools largely through the lens of opportunity such as leveraging digital skills, productivity and preparing the workforce of tomorrow. Norway has just spent its political capital on the opposite idea: restraint, phased carefully by age so that older teenagers are gradually trusted with the tools they will need, but younger children are protected from a shortcut that costs more than it appears to save.
That idea is no longer speculative. Researchers from Stockholm University and the University of Hong Kong tracked more than 26,000 secondary school students over thirty months and found, at first, an encouraging story. Homework grades rose by 18% while homework time fell by nearly a third – every student and parent’s dream. Then the researchers kept watching and it took roughly two years for the real cost to surface, hidden beneath a run of glowing homework grades. When those same students sat closed-book, supervised exams, scores were down by 20%. On the high-stakes exams that determine which course, which university and which future a teenager gets to pursue, the fall was as steep as 24%. Surprisingly, it was not the strugglers who suffered most but the strongest students, capable enough to prompt an AI into producing flawless homework, and in doing so, quietly forfeiting the cognitive struggle that is the actual mechanism by which knowledge lodges itself in a young mind.
Britain is no bystander to this story. Across UK universities this year, “traffic-light” assessment – red for strictly invigilated, amber for partial AI use, green for open encouragement – has spread from Bristol to Leeds to the London School of Economics. Institutions from Manchester to Glasgow to UCL’s law school have quietly rebuilt their exam halls, dusting off pen-and-paper exams because remote invigilating, it turns out, was never going to hold against a determined student with a second screen. The University of Bath is rolling out a formal “two-lane” model for the 2026–27 academic year, distinguishing closed, supervised assessment from open coursework in which AI is permitted but scrutinised. These are sensible reforms. But they remain a patchwork, university by university, without a shared national standard for what a genuine degree now requires.
Follow the same cohort forward, out of the lecture hall and into the labour market, and the reasons for urgency multiply. Adzuna’s job market report shows UK entry-level job postings for graduate roles, apprenticeships and internships have fallen by around a third since ChatGPT’s public launch in November 2022. Stanford University economists tracking 25 million American payroll records found a similar story on the other side of the Atlantic. Employment for 22 to 25-year-olds in AI-exposed occupations has fallen 13% since generative AI went mainstream, while employment for older workers in those same roles has grown by as much as 12%, insulated by a tacit, hard-won expertise no prompt can manufacture. Not every employer blames AI outright, and in the UK the causes are undeniably tangled with a sluggish economy and rising labour costs, but the direction of disruption – fewer rungs at the bottom of the ladder – is unmistakable, and it is exactly the ladder British graduates have historically climbed.
Put a generation already underperforming its predecessors together with a labour market quietly closing the door on entry-level roles, and the reality becomes uncomfortable. What happens to the cohort now coming through, raised not just on smartphones but on AI as a first instinct, when they arrive at that door only to find it narrower still, and employers with less patience than ever for candidates who cannot demonstrate genuine skills built on internalised knowledge?
I have spent enough time inside university corridors to know that many institutions have chosen, until very recently, to look the other way. This is understandable but it is also unsustainable. The harm runs deeper than any individual student’s lost agency. Picture the lecturer, marking a stack of essays she knows were not written by the students whose names sit atop them. Strip away detection as a solution and what remains is a harder, more fundamental question. Not who cheated, but what a degree is actually certifying.
Some UK institutions are answering it well. The University of Surrey’s move to redesign its entire curriculum by September 2026, assessing the process of learning rather than just the polished output, through looking at drafts, corrections and a student’s own critical account of where the AI went wrong, points to something more durable than a traffic light. But individual universities acting alone, however thoughtfully, cannot solve a sector-wide credibility problem.
What is needed is not a ban, at least not for university students on the cusp of a career, but assurance: a way of separating the graduates who did the cognitive work from those who outsourced it. That requires agreement that proof of competence in each learning outcome comes from secure, AI-free assessment: the supervised exam, the oral defence, the lab bench, the clinical placement, the encounters where knowledge has nowhere to hide. That assessment should act as a hurdle. Fail it, and the unit is failed, regardless of how fluent a student has become in the separate, complementary skill of using AI tools well.
An independent, internationally recognised certification, awarded annually to institutions that meet this bar, would hand British employers, parents and the students themselves something reputation alone has never offered at scale. Reputation only travels as far as the handful of universities an employer already happens to know, useful for Oxbridge and the red-bricks, far less so for the hundreds of institutions competing for talent in a graduate market that is now genuinely global and endlessly mobile. A certification, like the audit report on a company’s accounts, travels everywhere a CV does.
The question facing the UK is not only whether to follow Norway’s lead into the primary classroom. It is whether Britain acts now, at the point where degrees are conferred and careers begin, before the crisis Norway is trying to head off in its schools arrives, unaddressed, in universities, in the form of a generation of graduates who cannot demonstrate that what is written on their certificate is evidence of what actually resides in their heads. British industry cannot afford to retrospectively sort the genuine achievers from the graduates who rode the coattails of their favourite generative AI.
Dr. Alan Finkel
Alan Finkel AC is founder and executive chairman of Proudly Human. He was formerly Australia’s Chief Scientist, the Chancellor of Monash University, President of the Australian Academy of Engineering, founder of Stile Education and founder of Axon Instruments. He was the co-founder of popular science magazine Cosmos and is the author of Getting to Zero (2021) and Powering Up (2023).
