AI in higher education is no longer a question of policy preference. Since June 2024, when TEQSA required every provider to submit a credible, governance-approved action plan addressing the risk generative AI poses to award integrity, it has been a regulatory expectation. The regulator's position, set out in its 2023 principles on assessment reform and its 2025 guidance on enacting that reform, is that providers must both assure that graduates have achieved their learning outcomes without AI doing the work, and prepare those graduates to work in a world where AI is everywhere. A provider that cannot show it is doing both is exposed under Standards 1.4, 5.2 and 6.3.
That is the integrity question, and it is the one most written about. But it is only one of five AI considerations a provider's governing body should be able to speak to, and after fifteen years of TEQSA registration work I find the other four are usually the ones nobody has thought about. This article covers all five.
What has TEQSA actually said about AI in higher education?
Three documents matter. In 2023 TEQSA published Assessment reform for the age of artificial intelligence, a short set of principles and propositions developed with sector experts. Its two principles are that assessment and learning experiences should equip students to participate ethically and actively in a society pervaded by AI, and that forming trustworthy judgments about student learning in a time of AI requires multiple, inclusive and contextualised approaches. The propositions beneath them call for assessment that emphasises process and the development of judgment, that is secured at key points, and that is designed at the program level rather than unit by unit.
In June 2024 TEQSA issued its request for information on addressing the risk of AI, requiring each provider to lodge, by 3 July 2024, an institutional action plan overseen by its governance bodies. TEQSA said it would follow up with providers whose plans were absent, insufficient or unachievable, and would consider regulatory tools where a provider continued to fail to respond.
In 2025 it published Enacting assessment reform in a time of artificial intelligence, which moves from principle to practice, with examples of how institutions are restructuring assessment to address the risk to learning assurance while supporting students to use AI responsibly. None of these documents is a Threshold Standard, and the usual caveat applies: guidance is not law. But they tell a provider exactly how TEQSA will read Standards 1.4 on assessment and 5.2 on academic integrity at the next application, and that is what matters.
Consideration 1: award integrity and assessment design
The core regulatory concern is simple. If a student can complete the assessment tasks of a course using generative AI without achieving the learning outcomes, the provider cannot confirm that its graduates have those outcomes, and the qualification it certifies under Standard 1.5 is not what it claims to be. That is a risk to the integrity of the award itself, and TEQSA treats it as such.
The response TEQSA expects is assessment redesign rather than detection. AI detection software is unreliable, discriminates against some cohorts, and does not by itself confirm learning. The practical pattern that has emerged across the sector, and that TEQSA's 2025 guidance reflects, is program-level assurance: a small number of secured assessment points in each course, such as invigilated examinations, vivas, practical demonstrations or supervised work, at which the achievement of course learning outcomes is confirmed without AI, combined with assessment elsewhere in the course that assumes and teaches the use of AI as a professional tool. Some institutions describe this as a two-lane approach. The label matters less than the design: the academic board should be able to point to where, in each course, learning is assured, and to explain why those points are sufficient.
For a new provider this is straightforward to build in from the start, and TEQSA's reading of what it looks for in a new course now includes it. For an established provider it is a course review question, and the record of the academic board considering it belongs in the evidence for re-registration.
Consideration 2: the action plan and its governance
TEQSA's request was for a plan oversighted by the appropriate governance mechanisms. That phrase is doing a lot of work. A plan written by a learning and teaching manager and lodged without the academic board or governing body having considered it does not meet the expectation, and providers that took that shortcut in 2024 are now being asked to show what has happened since.
What TEQSA is looking for at the next application is the plan, the minutes of its approval, the evidence that its actions were implemented, and the academic board's ongoing monitoring of AI-related integrity risk as a standing item. In other words, AI belongs in the same self-assurance cycle as every other academic risk. Providers that treated the 2024 request as a one-off submission will find it returns as a question at renewal, and the academic board is where that question will be directed.
Consideration 3: graduate capability and the AQF
The second of TEQSA's 2023 principles is easy to overlook because it is not about risk. It says that students must be equipped to participate in a society pervaded by AI. That is a learning outcome question, and it connects to the AQF descriptors on skills and application: a graduate at level 7 or 9 who cannot use AI tools critically, ethically and effectively in their discipline is arguably not meeting the descriptor for professional practice in that discipline in 2026.
The implication for course design is that AI literacy, discipline-specific and taught rather than assumed, belongs in the learning outcomes and the curriculum, not only in the academic integrity policy. Providers that ban AI across the board are not only fighting a losing battle on integrity; they are producing graduates whose capabilities the market will read as dated. The AQF levels have not changed, but what it means to apply knowledge with initiative and judgment in professional practice has.
Consideration 4: AI in higher education operations
Beyond the classroom, AI in higher education operations is already widespread. Providers are using AI in admissions screening, in student support chatbots, in marketing copy, in the drafting of policies, in timetabling and, increasingly, in tutoring and feedback. Each of these engages a standard, and few providers have mapped which.
Admissions decisions made or shaped by AI must still be fair, transparent and capable of explanation under Standard 1.1, and a governing body should know whether an algorithm is screening applicants and on what basis. Student-facing chatbots are part of the support services under Standards 1.3 and 2.3, and a chatbot that gives wrong advice about progression rules or wellbeing services is a compliance failure.
Marketing copy generated by AI is still the provider's representation under Standard 7.1, and AI is prone to inventing accreditations, rankings and outcomes that the provider does not hold. Student data used to train or prompt AI tools engages Standard 7.3 on information management and the Privacy Act, and providers should know where that data goes.
AI-delivered teaching and feedback raise a further question under Standard 3.2. TEQSA's staffing standard assumes that academic judgment is exercised by qualified academic staff. A provider that substitutes AI for marking, feedback or tutoring at scale, without a qualified academic remaining responsible for the judgment, is thinning its academic workforce in a way that the risk indicators will eventually show and that an assessor will ask about.
Consideration 5: AI-written applications and the self-assurance model
The last consideration is the one I raise with every prospective client. TEQSA's move from Confirmed Evidence Tables to self-assurance coincided with the arrival of fluent, generic, AI-drafted applications, and in my view the two are connected. Assessors now read applications for the specific: the named staff, the actual facilities, the real decisions in the minutes, the provider's own account of why each standard matters to it. An application that reads well but describes no actual institution is precisely what the self-assurance model was designed to expose, and it draws more scrutiny rather than less.
That does not mean AI has no place in preparing an application. It is a capable drafting tool, and a provider that uses it to structure and polish material that its own people have written, from its own evidence, is doing nothing an assessor will object to. The line is authorship. The governing body and the academic board are non-delegably responsible for the application, and their responsibility is not discharged by a document they did not write, do not understand, and could not defend at a site visit. The same governance mistakes that stall applications apply when the ghost-writer is a model rather than a consultant.
What a governing body should be able to say
By way of summary, a well-governed provider's board should be able to answer five questions about AI in higher education without notice. Where in each course is learning assured without AI, and how do we know those points are enough? What did our 2024 AI action plan commit to, and what has happened since? Where in our curriculum do students learn to use AI in their discipline?
Then the operational two. Where are we using AI in our own operations, which standards does each use engage, and who is accountable? And can we defend every page of our last application as our own work?
A board that can answer those is ahead of most of the sector. A board that cannot has a year's work, and should start with the first question.
Download the Darlo Governance Readiness Checklist
— now including an AI governance section covering assessment assurance, the action plan, operational uses and application authorship, drawn from our TEQSA registration and governance work with private providers. Get the checklist
Want the full article?
Enter your email for free access to the rest of this guide and our TEQSA resource library.
Frequently asked questions
Does TEQSA require providers to have an AI policy?
TEQSA required every provider to submit a governance-approved action plan addressing the risk generative AI poses to award integrity in 2024, and it expects that plan to be implemented and monitored through academic governance. A policy alone is not sufficient; the plan, its approval and its implementation are what TEQSA reads.
Can providers use AI detection software to meet TEQSA's expectations?
Detection alone does not meet them. TEQSA's guidance emphasises assessment redesign, with secured points at which learning outcomes are confirmed without AI, and program-level assurance, rather than reliance on detection tools that are unreliable and do not confirm learning.
Is it acceptable to use AI to write a TEQSA application?
Using AI to draft and polish material the provider's own people have written from their own evidence is unobjectionable. An application generated by AI that the governing body did not author and cannot defend fails the non-delegation principle and is more likely to attract scrutiny under the self-assurance model.
Which Threshold Standards does AI engage?
Principally Standards 1.4 on assessment, 5.2 on academic integrity and 6.3 on academic governance, but also 1.1 on admissions, 1.3 and 2.3 on student support, 3.2 on staffing, 7.1 on representation and 7.3 on information management, depending on how AI is used in the provider's operations.
Dr Brendan Moloney is CEO of Darlo Higher Education, Australia's largest specialist TEQSA consultancy. He holds a PhD from the University of Melbourne, is a Cambridge University Press author on governance in higher education, and has advised private providers on registration and course accreditation for more than fifteen years.