By Stuart Kerr, Technology Correspondent, LiveAIWire
In December 2025, the Association of Chartered Certified Accountants announced that routine online exams would cease from March 2026. The CEO’s explanation was unambiguous: cheating technology had outpaced existing safeguards, undermining academic integrity within credentialing pipelines. What was being described is not a cheating problem. It is an assessment validity crisis: when the tools available to students can produce work that passes the tests institutions use to certify competence, the tests no longer certify what they are designed to measure. Higher education built its assessment infrastructure around tasks that AI is now exceptionally capable of performing. The question that follows from that fact is not how to prevent students from using AI. It is what a university education is actually for, what it should be teaching, and how it should be certifying that learning has occurred.
By 2025, 92 percent of UK undergraduates reported using generative AI tools for assessments, up from 66 percent in 2024, according to the Higher Education Policy Institute and Kortext survey. In the United States, 88 percent of college students reported using AI for assessments in some form. AI-related misconduct cases rose from 1.6 per 1,000 students in 2022 to 7.5 per 1,000 in 2026 — a 400 percent increase over three academic years. These numbers do not represent a moral failure by students. They represent a rational response to the availability of a tool that makes the existing form of assessment trivially easy to game, in an environment where institutions have not redesigned assessments fast enough to maintain the relationship between effort and demonstrated learning.
What AI Can Now Do in Academic Domains
The scope of AI capability in academic tasks has expanded substantially faster than higher education’s assessment frameworks have adapted. GPT-4 passed the Uniform Bar Examination with a score in the top 10 percent of human test-takers in 2023. Later models have extended that performance across professional certification exams in medicine, accounting, and finance. AI systems can now complete approximately 94 percent of theoretical computer science tasks, according to capability assessments conducted by AI labs and independent researchers. In essay writing, the benchmark for undergraduate-level work in most humanities and social science disciplines, multiple independent studies have found that AI-generated essays are rated as highly by instructors as human-written equivalents when the rater does not know which is which.
The implications for the credentialing function of higher education are serious. A university degree in law, accountancy, business, or the social sciences has historically certified that the graduate can perform specific analytical and written tasks at a defined standard. If AI can perform those tasks at the same standard, and if graduates can use AI to produce work that passes the degree’s assessment requirements without developing the underlying capability, the degree no longer certifies what it says it certifies. Employers who hire graduates based on their degree credentials are making an increasingly uncertain inference about the actual capabilities those credentials attest to.
How Forward-Thinking Institutions Are Responding
The institutions generating the most credible responses to AI’s challenge to higher education are those that have started with the question of what they are actually trying to certify, rather than with the question of how to stop students using AI. The answer to the first question is not the same across disciplines, but some common threads are visible. Higher education at its best certifies not the ability to produce a specific type of document under controlled conditions, but the ability to reason, synthesise, evaluate, and communicate in ways that reflect genuine engagement with a body of knowledge and a set of disciplinary practices.
That richer conception of what education certifies points toward assessment formats that AI cannot currently replicate: oral examination, in which the student must reason in real time in response to probing questions from an examiner who can follow unexpected threads; project-based assessment evaluated against criteria that require contextual knowledge specific to the course; portfolio assessment that tracks the development of thinking over time rather than measuring a point-in-time product; and clinical or practical placement assessment that observes performance in real-world application. These formats are more resource-intensive to design and administer than written examinations. They are also more authentic evaluations of the capabilities that higher education is supposed to produce.
The Credential Inflation Question
The widening availability of AI tools for academic work creates a credential inflation dynamic analogous to what happens to any certificate when it becomes easier to obtain. If the effort required to pass a degree falls because AI can do the work, and if institutions do not redesign their programmes to maintain the genuine challenge of learning, the degree’s signal value as an indicator of capability decreases. Employers who recruit graduates on the basis of degree credentials, and who are not themselves assessing AI capability separately from degree performance, are at risk of systematically hiring graduates who have less capability than their credentials suggest.
The PwC 2026 AI Jobs Barometer found that AI-exposed entry-level roles are seven times more likely to demand traditionally senior skills like leadership and strategic thinking compared to the least exposed roles. That finding has a corollary for higher education: the degree programmes that are most directly threatened by AI’s displacement of assessment tasks are those whose credentials most closely map to the tasks AI handles best. Programmes that certify the ability to produce standard-format essays, routine analysis, and codified problem-solving are most at risk of credential inflation. Programmes that certify the ability to exercise professional judgement, navigate genuine ambiguity, build and maintain client or colleague relationships, and synthesise knowledge across domains — capabilities that AI cannot replicate and that assessment formats can be redesigned to genuinely test — are most capable of maintaining their credential value in an AI environment.
The Return on Investment Question
The economic case for a university degree is being scrutinised more intensely in the AI era than at any previous point. The AI skills wage premium — 56 percent higher wages for roles requiring AI skills — has emerged faster through employer-delivered training and self-directed learning than through formal university programmes. Coding bootcamps, professional certification programmes, and online learning platforms have demonstrated that specific technical capabilities can be acquired in months rather than years, at costs a fraction of a university education. The question for prospective students is whether the combination of critical thinking development, professional network formation, disciplinary grounding, and credential signal that a university degree provides is worth the cost and opportunity time in an environment where specific skills can be acquired more efficiently elsewhere.
The honest answer is that it depends on the programme, the institution, and the individual’s goals in ways that are more important and more variable than they were before AI expanded the alternative pathways to professional capability. The degrees that will maintain their return on investment are those that genuinely develop the capabilities that neither AI nor cheaper alternatives can replicate: deep disciplinary reasoning, professional judgement, ethical practice, and the interpersonal capabilities required to lead complex organisations. The degrees that will struggle to justify their cost are those that primarily certify task performance that AI has made automated and therefore less scarce.
For readers navigating higher education decisions in the AI era, LiveAIWire’s coverage of what the academic integrity data reveals and our analysis of AI tutoring and the attainment gap provides the evidence base. The question of the digital literacy deficit addresses what education systems should be building alongside AI tool familiarity.
The Assessment Reform That Is Actually Working
The institutions generating the most credible responses to AI’s challenge to higher education are those that have started from the question of what they are actually trying to certify, rather than from the question of how to stop students using AI. The answer points toward assessment formats that AI cannot currently replicate: oral examination in which the student must reason in real time in response to probing questions; project-based assessment evaluated against criteria that require contextual knowledge specific to the course; portfolio assessment that tracks development of thinking over time; and practical placement assessment that observes performance in real-world application. In the UK, 59 percent of undergraduates said the way they are assessed has changed significantly because of generative AI. The proportion of students saying university staff are well-equipped to work with AI doubled in twelve months, from 18 percent in 2024 to 42 percent in 2025. That is genuine progress. It is also evidence that 58 percent of students still do not believe their institutions are ready.
The institutions that are moving fastest on assessment reform share a characteristic: they have made the redesign a strategic priority rather than a departmental problem. When assessment redesign is left to individual instructors managing increased workloads in the absence of institutional support and resource, the result is uneven and inadequate. When institutions invest in professional development for assessment design, provide additional resources for oral examination and portfolio review, and create shared frameworks for AI-appropriate assessment across programmes, the pace of reform becomes adequate to the pace of change. The technology will continue to improve. The institutions that build assessment frameworks robust to further AI capability improvements — rather than frameworks designed to catch the capabilities of 2025 models — are the ones that will maintain the credibility of the credentials they confer.
What Employers Are Starting to Do
The employer response to credential uncertainty is beginning to shift the hiring landscape in ways that reinforce the pressure on universities. Skills-based hiring — evaluating candidates on demonstrated capability rather than degree credentials — has been growing as a trend for several years. AI’s disruption of assessment validity is accelerating it. McKinsey’s research on AI and employment found that employers in AI-exposed sectors are increasingly requiring portfolio demonstrations, practical assessments, and case studies rather than relying primarily on degree class as a signal of capability. IBM, Google, and Apple have notably removed degree requirements for specific roles. The broader shift from credentialing to demonstrated competence as the primary hiring criterion is not complete and may not become dominant across all sectors, but it is significant enough to change the calculus for prospective students evaluating the return on investment of specific degree programmes.
The Global Dimension and What It Means for Access
The AI challenge to university credentials is not uniformly distributed globally, and the responses available to different national higher education systems vary enormously based on resource levels. Elite universities in high-income countries have the resources to redesign assessment, invest in oral examination infrastructure, and develop AI literacy programmes that prepare graduates for AI-augmented professional environments. Universities in middle-income and lower-income countries, which educate the majority of the world’s university students, face the same challenge without the resources to respond at equivalent pace or scale. The result is a risk that AI disrupts credential value most severely in the institutions least equipped to respond — institutions whose graduates are already at a competitive disadvantage in global labour markets that increasingly reward credentials from elite institutions in high-income countries.
The counter-argument is that AI tools are also equalising access to educational quality in ways that benefit students in lower-resourced institutions. An AI tutoring system available on a smartphone can provide personalised learning support at a level that only the best-resourced institutions could previously offer through human tutoring programmes. The same AI that disrupts the validity of essay-based assessment can also, when used as a genuine learning tool rather than an assessment substitute, extend access to the kind of Socratic intellectual engagement that has historically been the privilege of students at institutions with small class sizes and engaged faculty. Whether AI in higher education net-widens or net-narrows educational equity is an empirical question that depends on which effects dominate — and that question will be answered differently in different institutional contexts and national systems over the next decade.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and everyday life. LiveAIWire publishes original AI journalism every weekday at liveaiwire.com.