The cohort arriving in 2027 will not reason like the one in front of you. Are universities ready?

Join HEPI for an ‘In Conversation’ webinar ahead of the 2026 admissions round on 11:00-12:00, 11th August 2026, featuring Dr Jo Saxton CBE, Chief Executive of UCAS. In discussion with Nick Hillman OBE, followed by a live Q&A, the session will explore the key issues shaping this year’s admissions cycle and what they mean for universities, applicants and the wider higher education sector. Register now.

This blog was kindly authored by Roney Lima do Nascimento, an IB Diploma Programme mathematics teacher, and author of Generative AI for Teachers.

Every university plans its teaching around the students it can see. That is the problem. The undergraduates in seminar rooms today formed their study habits, their relationship with sources and their independent reasoning before generative artificial intelligence reached classroom scale. The cohort that will matriculate from 2027 onwards is different in kind, not degree. These students will arrive having spent the whole of their secondary education alongside capable AI and the scaffolding they bring will have been shaped by a decision their schools made, often without realising they were making it.

I have spent two years training teachers in AI literacy and studying how schools actually absorb these tools, alongside teaching IB Diploma mathematics to students who feed directly into UK higher education. The pattern that matters for universities is this: the same tool, under the same training, produces opposite cognitive outcomes depending on the posture the institution takes towards it. Schools that framed AI as a competence to develop – where disclosure was safe and use was taught – are producing students with measurably stronger AI-augmented reasoning. Schools that defaulted to a policing posture, where AI was treated as a transgression to detect and punish, are producing students who used the tools anyway, in secret, without ever being taught to interrogate what the tools gave them. The first group learned to think with AI. The second group learned to hide that they were not thinking at all.

Both groups will arrive on UK campuses in 2027, under the same admissions framework, often in the same seminar rooms. Universities will not be able to tell them apart from a personal statement or an A-level grade. The difference will surface in the first term, when one student uses AI to pressure-test an argument and deepen it and the next uses AI to manufacture an argument they cannot themselves defend. Same grade on entry, opposite intellectual readiness.

The temptation for the sector is to manage this through detection. That instinct is understandable and almost entirely wrong. Detection tools are an arms race the institution cannot win and worse, they teach exactly the lesson universities least want to teach, that the goal is to avoid being caught rather than to reason well. A policing posture at university scale will simply reproduce, in higher education, the concealment that the weaker schools produced. The students who most need to be taught how to think with these tools will be the ones with the strongest incentive to hide that they cannot.

There is a better question for the sector to ask, and it is a question of institutional posture, not technology procurement. What does a university actually want its graduates to be able to do with AI by the time they leave, and is the institution teaching that explicitly or assuming it will happen by osmosis? The honest answer at most institutions today is that the posture has never been decided. Many institutions will, rightly, note that they are already treating AI as a curriculum question and not merely an IT or integrity matter. The posture question sits underneath that work. It is not which modules use AI tools, but what kind of reasoning the institution recognises as a graduate-level performance once AI is available in the room. That is a question upstream of curriculum design, and at most institutions, it has not been answered explicitly.

Three things follow for universities preparing for the 2027 discontinuity, and none of them require new spending.

• First, diagnose on entry rather than assume. A short, low-stakes reasoning task in the first weeks, designed to reveal whether a student can use AI to strengthen an argument or only to generate one, tells an admissions tutor far more than a transcript.
• Second, decide the posture explicitly and write it into how teaching, not just integrity, is designed. A department that can articulate what competent AI use looks like in its discipline can teach it; one that only knows what it wants to forbid cannot.
• Third, treat the gap between the two incoming cohorts as a teaching problem to be closed in the first year, not a fixed property of the student. The weaker-prepared students are not less able. They were taught by institutions that made the wrong posture decision, and a university that makes the right one can still reach them.
  
  The cohort arriving in 2027 is the first of many. Every subsequent intake will have spent longer alongside these tools, and the divergence between the well-taught and the merely-exposed will widen, not narrow. Universities that decide their posture now, while the first such cohort is still a year away, will be teaching deliberately when those students arrive. Universities that wait will be reacting to a divergence they could have seen coming, and reaching for detection tools that teach the wrong lesson to the students who can least afford to learn it.
  
  The decision is not whether to allow AI. That decision has already been made, in every secondary school in the country, with or without a policy. The decision in front of universities is whether to teach, on purpose, the reasoning their degrees are supposed to certify, or to assume a generation will acquire it on its own. The schools that assumed are already producing the students who did not.