Rethinking AI in Accounting Education
This article will count 0.25 units (15 minutes) of unverifiable CPD. Remember to log these units under your membership profile.
A blueprint for higher education institutions
Guidance from CIBA for universities, colleges and private providers offering accounting qualifications
In July 2026 the University of Chicago Law School published a strategy for teaching in the AI era. It is worth the attention of anyone who educates accountants, because it answers a question every accounting programme now faces: what does a qualification prove when a generative tool can produce the work?
Chicago’s answer is to stop treating AI as a single problem with a single rule. Instead it differentiates by the stage of learning. For beginners it protects the effortful struggle that builds competence. For advanced work it teaches students to use AI well. And it introduced a requirement that translates directly into accounting: before a major piece of work counts, the student must defend it out loud, answering questions that reveal whether the thinking is their own.
Some of Chicago’s specific measures suit a residential law school and will not fit every accounting provider. Distance, online and part-time programmes cannot ban devices, and should not try. The value is in the logic, not the mechanics. This blueprint takes that logic and sets out what accounting programmes should do to rethink AI, whatever their mode of delivery.
One point should be clear from the start. AI did not create the problem it is now blamed for. South African employers were already reporting that graduates arrive able to reproduce theory but not to apply it, and the Pnet data shows the issue is a mismatch between what qualifications certify and what work requires, rather than a shortage of qualified people. AI simply made that gap impossible to ignore, because a tool can now produce the polished output that used to stand in for competence. The redesign set out here therefore does two jobs at once: it makes assessment credible in the AI era, and it closes the work-readiness gap employers have flagged for years. That is why it is worth the effort even for a programme that feels its curriculum is already full.
A note on terms. This blueprint is about AI, and generative AI in particular, not the broader “digital” shift of cloud software and automation that accounting programmes have largely absorbed. Cloud tools did not threaten the integrity of assessment. Generative AI does, and it is the specific challenge addressed here.
Start from the qualification, not the technology
The purpose of an accounting qualification is to certify that a graduate can do defined work to a professional standard. AI changes what a given assessment proves, but it does not change that purpose. The design question is therefore specific: for each competency, does the assessment still test whether the graduate can do the work, or only whether they can obtain an answer?
That question sorts a curriculum into three kinds of learning, and each calls for a different response.
Build core competence AI-free in the foundation phase
Early accounting modules teach the mechanics on which everything later depends: the accounting equation, double entry, the preparation of financial statements, basic tax computations, reconciliations. A student who lets AI carry this stage will not have the foundation to judge AI’s output later, when the stakes are higher and the errors are harder to see.
For these foundational competencies, assessment should be conducted under conditions the student cannot outsource. In contact settings this means invigilated, closed-tool assessment. In distance and online settings it means invigilated online proctoring, timed problem-solving that resists copy-and-paste answers, or short oral checks by video. The mode varies. The principle does not: at the foundation stage, the programme must confirm the student can perform the core mechanics unaided.
This is not nostalgia for pen and paper. It is the same reasoning a practitioner uses when they will not sign off figures they cannot themselves reconstruct.
Teach AI as a competency in its own right
The mistake to avoid is treating AI either as forbidden or as something students will simply pick up. Neither is adequate for a profession that already runs on these tools. Accounting programmes should teach the use of AI explicitly, as assessable content, and the most effective structure mirrors Chicago’s approach to writing: do the work without AI first, then layer AI on top.
In practice this means a student prepares a set of financial statements or a tax computation unaided, then uses AI to draft the supporting notes, to test the treatment of a transaction, or to critique their own work, and then reviews the tool’s output against their own. The competency being assessed is not the AI’s answer. It is the student’s ability to prompt the tool, to spot where it is wrong, to correct it, and to explain the professional and legal reasons for the correction, including obligations under the Protection of Personal Information Act. A graduate who can supervise AI is employable. One who can only trust it is a liability.
Shift advanced assessment to what AI cannot own
As students move to advanced modules and electives, the useful work is exactly the work AI struggles with: interpreting results, advising a client, exercising professional scepticism, weighing an ethical conflict, defending a position under challenge. Here the programme can and should allow supervised AI use, because the assessment is no longer about producing an output that a tool can generate.
This is also where programmes should experiment. Case studies where students critique an AI-produced analysis, assessments that require a recommendation rather than a calculation, and structured oral examinations all test higher-order judgement that a generative tool cannot fake. The goal is assessment that becomes more revealing, not less, when the student has AI available.
Make the oral defence a standard feature
The single most transferable idea from Chicago, and the cheapest to implement, is the oral defence. Any substantial piece of work, a research report, a capstone project, a set of prepared financial statements, a tax opinion, should be followed by a short discussion in which the student answers questions about their reasoning.
For accounting this is not an artificial exercise. It rehearses what practitioners do constantly: explain the numbers to a client, justify a treatment to a reviewer, defend a position to SARS or a bank. An oral defence confirms authorship without banning the tools that helped produce the work, and it develops a skill the graduate will use for their entire career. Programmes should build it into major assessments across the qualification, not reserve it for the final year.
Build the reasoning the assessment protects
Assessment that demands reasoning is only fair if the programme has taught reasoning. Critical thinking, sound reasoning and professional judgement are the capacities the AI era rewards most, and they must be developed deliberately rather than assumed.
The evidence points to a clear method: reasoning improves most when its principles are taught explicitly, by name, and then embedded in deep subject content and applied repeatedly, a finding from the meta-analytic work of Abrami and colleagues and from Willingham. Because reasoning draws on secure domain knowledge, a student reasons well about a going-concern judgement or a deferred tax position only when they know the accounting deeply. The reasoning core therefore belongs inside the discipline. In practice that means teaching the moves of accounting reasoning explicitly and modelling them on worked examples, then having students practise on progressively harder problems with the scaffolding removed, and requiring them to explain and defend their thinking aloud. Case-based and problem-based learning is the natural vehicle.
Philosophy has a valuable part to play, incorporated through accounting cases. Three strands are directly useful to an accountant: logic, for recognising a valid argument and a fallacy, which is exactly what evaluating an AI-generated answer requires; epistemology, the discipline of asking how we know a claim is warranted and what counts as evidence, which is the core of professional scepticism; and ethics, taught as reasoning through real dilemmas rather than as a list of rules. Woven into the accounting curriculum, these develop the single capacity the AI era most rewards: judging whether a claim can be trusted.
Redesign assessment rather than police it
South African institutions have already learned that AI-detection software does not solve this. The University of Cape Town and Stellenbosch University have withdrawn AI detectors as unreliable, and the evidence on false positives, particularly against students who write in English as an additional language, makes them unsafe to rely on for high-stakes decisions. The Council on Higher Education’s discussion volume on AI, Kagisano 15, and the growing register of institutional policies maintained by Universities South Africa point the same way: the durable response is to redesign assessment, not to hunt for the machine.
It is worth being clear about where national guidance stands, because programmes should not wait for it. Government has published a Draft National AI Policy (Government Gazette 54477, 10 April 2026), still in consultation and led by the Department of Communications and Digital Technologies. It treats AI education as a skills priority but does not address assessment or academic integrity, and national AI policy guidelines are only expected in 2026/27. The Council on Higher Education’s Kagisano 15 is a discussion volume, not a standard. Guidance specific to higher education assessment therefore remains, for now, a matter for institutions and professional bodies.
The stakes are not hypothetical. South Africa’s own draft national AI policy was reported to have cited research that did not exist, generated by AI and never checked. That is the failure this blueprint is designed to prevent, appearing at the level of national policy: an output that looked authoritative, with no competent human standing behind it. An accounting qualification exists precisely to guarantee that a competent human stands behind the work.
What good looks like across a programme
A well-designed accounting qualification should now show a clear progression. In the foundation phase, students demonstrate core competence without AI. In the middle phase, they learn to use AI and to supervise it, assessed on their judgement of its output. In the advanced phase, they work with AI on higher-order tasks while defending their reasoning in person. Read end to end, the programme teaches students to work without, with, and about AI, in that order.
Three institutional enablers hold this together:
Every module should state its AI position explicitly and communicate it to students, so expectations are clear rather than left to individual discretion.
Academic staff need development and time to redesign assessment, because this cannot be delegated to a policy document.
The programme should review its approach regularly, because the tools and the legislation will keep changing.
The bottom line for accounting educators
Chicago’s contribution is not a set of rules to copy. It is a way of thinking: match the response to the stage of learning, protect competence where it is formed, teach AI where it is needed, and test the human directly where it matters. For accounting programmes, that thinking leads to assessment a student cannot outsource in the foundation phase, explicit instruction in supervised AI use through the middle of the qualification, and an oral defence of substantial work throughout.
The through-line is simple, and it is the same answer to AI and to employability. The profession’s protected asset is judgement, the ability to decide whether a claim is warranted and to stand behind it. Education’s job is to build that judgement and to prove it, rather than to certify polished output that a machine can now produce. A qualification that does this remains what employers, clients and professional bodies need it to be: reliable evidence that the graduate can do the work. CIBA offers this blueprint to its education partners in that spirit, and welcomes the chance to work through it with any institution rethinking how it teaches and assesses accounting in the AI era.
References
● University of Chicago Law School, “Rethinking Legal Education in the AI Era” (9 July 2026).
● Council on Higher Education, Kagisano 15: Artificial Intelligence and Higher Education in South Africa (discussion volume).
● University of Cape Town, “UCT scraps flawed AI detectors” (2025).
● Universities South Africa, register of institutional AI policies and guidelines.
● Draft South Africa National Artificial Intelligence Policy, Government Gazette 54477 (10 April 2026).
● Abrami et al., “Strategies for Teaching Students to Think Critically: A Meta-Analysis,” Review of Educational Research (2015).
● Willingham, “How Can Educators Teach Critical Thinking?” American Educator (2020).