From permissions to purpose: a program level approach to assessment in an AI-integrated university
By A/Prof. Tracy Wilcox, A/Prof. Mark Ian Jones and Prof. Alison Beavis
Published on 12 August 2026
For the past two years, much of higher education's conversation about generative AI has centred on a deceptively simple question:
What should students be allowed to use?
Universities responded with AI categories, traffic light systems and lists of permitted or prohibited tools. These frameworks were an understandable first response to a rapidly evolving technology and a genuine attempt to preserve academic integrity while institutions found their footing.
But AI continues to accelerate.
Generative AI is no longer an emerging novelty. It is becoming part of everyday professional practice across disciplines, from engineering and business to education, law, health and the creative industries. As AI becomes embedded in the way people work, the question facing universities also needs to evolve.
Recent sector thinking, including the Castlereagh Statement, suggests that the central educational question is no longer what technologies students use, but which enduring human capabilities universities should value, develop and assure in an AI-enabled world. The statement argues that the challenge for universities is no longer what students know alone, but how they develop judgement, adaptability, ethical reasoning and other distinctly human capabilities that remain essential in an AI-enabled society.
Rather than asking what tools are allowed, we should instead ask:
What kind of learning are we trying to assure?
This conversation is unfolding across the sector – locally and globally. At UNSW, we announced a new three-mode assessment approach during our new Assessment Policy Town Hall on 22 June 2026, as part of a broader discussion about moving beyond permissions-based approaches to AI in assessment. The development of these assessment modes was undertaken with a clear line of sight to the national and international conversations shaping the future of AI in higher education, including the work being led by TEQSA. Four days later, TEQSA released Assuring Quality Learning in a Gen AI-Integrated Future. The close alignment between the two approaches was encouraging, reflecting a shared direction across the sector. Both move away from regulating AI as a technology and instead focus on assuring learning through the deliberate development and assessment of graduate capabilities in an AI-enabled world.
Assessment must be driven by educational purpose
One unintended consequence of permissions-based AI frameworks is that they can shift attention away from learning itself.
Simply specifying whether AI is permitted tells students little about what they are expected to learn, what capabilities they should demonstrate, or why an assessment has been designed in a particular way. Permissions regulate tools and their use. They do not define educational purpose.
Assessment categories should instead communicate the capability being demonstrated. They should help students understand the purpose of a task and help educators design assessment that provides meaningful evidence of learning. This reflects arguments across the sector that assessment reform must move beyond AI detection towards authentic verification of capability at key points across a program.
Assessment categories should therefore communicate why a task has been designed in a particular way, not simply whether a technology may be used.
As AI becomes part of the educational landscape, assessment increasingly needs to distinguish between three educational purposes:
- applying knowledge in authentic professional contexts
- demonstrating independent disciplinary capability
- developing and demonstrating capability in the responsible use of AI
These are fundamentally different learning goals, even if similar technologies are involved.
This educational shift is reinforced by TEQSA's guidance and the Castlereagh Statement, which argue that universities should focus on assuring learning by developing students' adaptive capabilities, including evaluative judgement, critical thinking and ethical reasoning, in assessment environments where generative AI is now an enduring feature. The emphasis moves beyond controlling technology towards designing assessment that provides credible evidence of learning while preparing graduates for an AI-integrated future.
A program design decision, not an isolated course decision
Perhaps the most significant shift is recognising that these assessment modes are fundamentally a program architecture decision rather than a series of course-by-course choices. Program teams determine how assessment modes are distributed across the curriculum to ensure students progressively develop disciplinary knowledge, AI capability and academic integrity in a coherent, scaffolded way. Within that shared framework, course convenors continue to exercise their disciplinary expertise by designing authentic assessment tasks that align with the program’s agreed assessment strategy.
The purpose of the framework is to ensure that every graduate has appropriate opportunities across their degree to demonstrate independent achievement, engage in authentic professional practice and develop AI capability.
Like communication, teamwork and ethical reasoning, AI capability develops progressively and therefore requires intentional scaffolding across a program. Likewise, assurance of independent achievement is strongest when it is planned across the curriculum rather than relying on disconnected course-level decisions. The balance between these assessment modes should therefore be designed deliberately at program level, with individual courses contributing to a coherent assessment strategy.
Within this architecture, the three assessment modes each serve a distinct educational purpose. These student-facing assessment modes deliberately answer the practical question students ask: Can I use AI for this assessment? Behind them sits a program architecture that answers a different question: Why has this assessment been designed this way?
Can use AI
Most university assessment prepares students for professional practice.
In contemporary workplaces, professionals routinely use a wide range of resources, technologies and increasingly AI tools. These assessments recognise that reality. Students may use AI because AI is simply one resource among many available to contemporary professionals. Assessment focuses on disciplinary understanding and professional judgement, not on the use of AI itself. AI supports learning; it does not replace it.
The focus shifts from asking Did you use AI? to Did you use it appropriately, responsibly and in ways that supported the intended learning outcomes?
Can't use AI
Some disciplinary capabilities must be demonstrated independently.
These assessments provide assurance that students have independently achieved key Program Learning Outcomes by providing robust evidence of individual knowledge, judgement and capability without reliance on AI or other unauthorised assistance. Typical examples include supervised examinations, interactive oral assessments, practical demonstrations, laboratory work and other controlled assessment environments.
These assessments remain essential for assuring graduate capability, meeting accreditation requirements and maintaining confidence in university qualifications.
Must use AI
Some assessments explicitly require students to use AI because AI capability is itself a learning outcome.
Students are expected to demonstrate critical, ethical, responsible and effective use of AI because these capabilities are increasingly expected in professional practice. Assessment focuses not simply on producing work with AI, but on evaluating AI outputs, recognising bias and hallucinations, understanding ethical and legal implications, and exercising informed professional judgement.
Importantly, universities must distinguish between using AI and understanding AI.
Some students, and indeed some academics, may choose to be conscientious objectors to the routine use of generative AI, whether for ethical, environmental, disciplinary or personal reasons. Universities should respect and invite scholarly debate around these issues wherever they are compatible with learning outcomes.
However, personal non-use does not remove the professional obligation to understand AI's capabilities, limitations and implications. Every graduate, regardless of whether they choose to use AI in their own practice, should be able to critically evaluate AI-generated outputs, recognise bias and misinformation, understand ethical, legal and sustainability considerations, and make informed judgements about when AI should, and should not, be used.
AI literacy is not synonymous with AI use. It is the capacity to exercise informed professional judgement about AI, including the judgement not to rely on it.
This position is consistent with the Castlereagh Statement’s call for discipline-specific AI literacy that includes ethical responsibilities. Graduates need not become uncritical adopters of AI, but they should understand its application, risks, limitations and implications within their own professional fields.
The Castlereagh Statement also highlights the growing importance of relational teaching. As information access becomes increasingly mediated by AI, high-value human interactions, feedback, mentoring, dialogue and professional judgement become even more important components of the university experience. Technology should serve pedagogy and public values, not determine them.
A framework designed to endure
Technology will continue to change. The AI tools students use five years from now will almost certainly be different from those available today. An assessment framework built around specific technologies or lists of permitted tools risks becoming outdated almost as quickly as it is published. A framework grounded in educational purpose is far more resilient.
The three assessment modes are not primarily a framework for AI. They are a framework for assuring learning in an AI-enabled university. They are about the educational purposes that universities have always been responsible for. Preparing students for authentic professional practice. Assuring independent achievement of Program Learning Outcomes. Developing graduates who can exercise sound judgement in a world where AI is ubiquitous.
Ultimately, the goal is not to regulate AI. It is to assure learning while preparing graduates for the realities of contemporary professional practice.
That means moving beyond permissions and prohibitions towards assessment designed around educational purpose, program architecture and the capabilities graduates will need throughout their careers.