06 August 2026
The freshers arriving at universities this autumn have something in common with no previous cohort: they began secondary school in the year that ChatGPT was released to the public. They are, in a meaningful sense, the first genuinely AI-native university students: the first cohort for whom AI assistance has been a normal feature of academic life since the beginning of their secondary education.
This is not a trivial distinction. Previous cohorts encountered AI tools as an intrusion into established habits: something they had to decide whether to use, something that sat uneasily alongside the norms they had learned. For these students, AI was simply part of the environment. They did not choose to integrate it; they grew up inside it.
What does this mean for universities? The short answer is that institutions are only beginning to work out the implications. The honest answer is that much of the received wisdom about how students learn, how they engage with knowledge and how they need to be taught is going to have to be revisited.
Consider the relationship with text. Students in earlier cohorts learned to read and write in environments where text was the primary vehicle for knowledge transmission, and where the ability to produce coherent written argument was a core academic skill developed over years. For AI-native students, text remains important, but its relationship to their own cognitive processes has changed. They have grown up routinely externalising writing tasks to AI tools, using them not to cheat in the traditional sense but to manage the cognitive load of academic work. The distinction between their thinking and AI-generated text is, for many of them, genuinely blurry in a way that it simply was not for previous cohorts.
This creates a significant challenge for universities, which have built much of their assessment infrastructure on the assumption that written work represents the student’s own thinking. This assumption was always somewhat imperfect – students have always been supported in various ways by others – but it is now substantially broken. The question is not how to enforce the old norm but how to design assessment in ways that are actually valid: that measure the things that matter in an environment where AI assistance is a given.
The institutions that are responding most thoughtfully are not trying to eliminate AI from academic work. They recognise that attempting to do so would be futile and would also deprive students of tools they will use throughout their professional lives. Instead, they are redesigning assessment toward the kinds of performance that AI cannot replicate or that require human presence and judgment: oral examination, live problem-solving, portfolio assessment that documents development over time, project work with real-world stakeholders who can verify the student’s contribution.
These are not new ideas. Educators have been advocating for more authentic, performance-based assessment for decades. What AI has done is make the case for these approaches urgent in a way that purely pedagogical arguments never quite managed. When the alternative is assessment that is trivially gameable by AI tools, the case for assessment reform becomes hard to resist.
The curriculum implications are less clear. There is genuine uncertainty about what knowledge and skills will be most valuable for this cohort, given the pace of AI development. It is possible to identify some relatively safe bets: the ability to evaluate AI-generated content critically, the capacity to identify when AI tools are likely to produce errors or biases, the ethical and governance dimensions of AI deployment. But beyond these, the picture is murky. The skills that seem most important today may be automated within years.
What seems more durable is a set of capacities that are genuinely difficult to automate: the ability to formulate the right question, to exercise judgment in contexts of genuine uncertainty, to navigate ethical complexity, to build and maintain trust in relationships. These are the capacities that the best university education has always tried to develop. The difference is that they now need to be foregrounded explicitly, because the alternative – teaching students to do things that AI does better – is no longer defensible.
The AI-native cohort is also arriving with different social and psychological characteristics that universities are still working to understand. There is growing evidence of higher rates of anxiety and social difficulty, of reduced tolerance for ambiguity, of a different relationship with failure and uncertainty. Some of this may reflect the broader mental health trends that have been visible for a decade. Some may be specifically connected to growing up with AI tools that resolve uncertainty immediately, that provide answers without requiring the discomfort of sustained inquiry.
Universities are going to have to develop new ways of thinking about student development that take these characteristics seriously, neither pathologising them nor accommodating them uncritically. The challenge is to help students develop the capacities that extended human development produces – patience, tolerance for ambiguity, the ability to sit with uncertainty – without simply reimposing the norms of previous generations who grew up in a very different information environment.
None of this is straightforward. Universities are already under significant pressure: financial, political, demographic. Adding a fundamental rethinking of pedagogy and curriculum to that list is genuinely daunting. But the alternative – continuing to run institutions designed for a previous cohort with different characteristics and different needs – is worse. The AI-native students arriving this autumn deserve universities that are actually designed for them, not for the students who came before.
.jpg)
