Secure and open assessment
How Australia's regulator and three universities are redesigning assessment for generative AI
IAPQB Report No. 11 · Practice report · Published 17 November 2026
International AI Practice Qualification Board (IAPQB)
In one minute
- Since 2023, papers published by Australia's higher education regulator, TEQSA, have said AI use cannot be reliably detected [1, 2]. The 2023 paper proposed securing key assessment points across each degree [1].
- The University of Sydney sorts tasks into two lanes: supervised secure tasks, and open tasks in which AI use cannot be banned or restricted. The change runs from 2025 into 2027 [3, 4, 5, 6].
- From 2027, the University of Queensland (UQ) will require a minimum share of secure assessment in every course, and UNSW Sydney a secure assessment for every learning outcome of a degree [7, 8, 9].
- A TEQSA toolkit case study says marking should follow the learning a task is meant to show [10]. At UQ and UNSW, students are told in writing what AI use is allowed [7, 9]. Secure tasks take staff time [2, 4].
1. A familiar situation
A lecturer is writing the brief for the main take-home essay in a second-year subject. A chatbot can produce a passable draft in a minute, and no tool can reliably show whether a student used one. A line saying "no AI" would take seconds to add. Nobody could check it.
2. Why universities split assessment in two
In a take-home task, the marker sees only the finished work. TEQSA's 2023 paper says detecting AI use reliably appears almost impossible [1]. Its 2025 guide calls detection with certainty "all but impossible" [2].
A rule in the brief does not change this. A 2025 peer-reviewed paper calls such rules discursive changes: instructions that students can ignore. It says they create an "enforcement illusion" [11]. TEQSA's 2025 guide adds that, in most cases, allowing any AI use in a task in effect allows it for the whole task [2]. The other route is to secure the task, for example by supervision. TEQSA's 2023 paper says such security is often resource-intensive, so it is not feasible for all assessments [1].
IAPQB view. The three universities in section 3 define a secure task by how it is run: it must be supervised, and at UQ and UNSW the student must also be identified [5, 7, 9]. All three are bringing in their rules in stages, with steps still planned for 2027 and, at UNSW, 2028 [6, 7, 8, 9].
3. Four cases
3.1 Australia, 2023 to 2026: the regulator's papers
An expert paper commissioned by TEQSA, dated November 2023, proposes "security at meaningful points across a program to inform decisions about progression and completion" [1].
In June 2024 TEQSA asked all Australian providers for an action plan on the risk generative AI poses to award integrity; all responded [12]. In September 2025 it set out three pathways: assure learning across the whole degree; put at least one secure task, which students must complete to pass, in every subject; or combine the two [2]. A June 2026 paper commissioned by TEQSA says secure assessment is still needed to confirm the required learning [13].
3.2 Sydney, 2023 to 2027: two lanes become policy
A post dated July 2023 on the University of Sydney's Teaching@Sydney blog proposed the two lanes [14]. In November 2024 the university announced that from Semester 1 2025, AI use would be allowed by default, except in exams and in-semester tests or where teaching staff decided otherwise. Its Pro Vice-Chancellor (Educational Innovation) said a blanket ban was "not practical, enforceable or desirable" [3].
The two lanes, with new assessment categories, were to apply from Semester 2 2025 [3, 4, 5]. Secure tasks are supervised in person, such as exams, tests, interactive orals and observed placements. In open tasks, AI use "cannot be banned, restricted, or controlled" [5]. An October 2025 post says all existing assessments for 2025 have been mapped to the framework, 2026 is for redesign, and by 2027 all online programmes are likely to need in-person assessment [6]. In 2024, around 80% of Sydney's units, with just under 60% of enrolments, had no exam [4].
3.3 Queensland, 2025 to 2027: a minimum share of secure marks
UQ defines secure assessment as work done "by identified students under supervision". Open assessment is done without supervision, and students may use AI. UQ turned off Turnitin's AI detector in mid-2025 [7].
From 1 January 2027, every course (a single subject) must include secure assessment worth at least 60% of the grade, or a secure task worth at least 30% that students must pass to pass the course [8]. Open assessment must expect AI use. An Associate Dean (Academic) may waive the rule for up to three years, only where secure assessment is not feasible and security elsewhere in the degree still shows that students did their own work [8].
3.4 UNSW Sydney, 2023 to 2027: from levels of AI use to secure checks
From February 2023 UNSW recommended categories of permitted AI use for each task, six of them by July 2024, when its Director of AI Strategy (Education) called two lanes "a great metaphor, but reality is more nuanced" [15]. Its staff guidance, updated in November 2025, still lists the six categories [16].
UNSW's Assessment and Progression Policy took effect on 1 June 2026. From 1 January 2027, each learning outcome of a degree or specialisation must be assured by at least one secure assessment, which uses identity verification and managed settings [9].
4. How practitioners handle it now
Secure points are chosen across a whole degree. Sydney's framework says secure tasks should be planned at degree level, at milestone and summative points, with an assessment map for each degree [5]. UNSW will require a secure check for every degree outcome [9].
A secure task is supervised, and the student is identified. At UQ, secure tasks are done by identified students under supervision, which can be invigilated, observed or mentored, including on placement [7, 8]. UNSW lists observed lab work, interactive oral presentations, placements, invigilated exams and oral assessments [9].
Open tasks assume AI use, and marking follows the purpose of the task. Sydney's FAQ says rubrics that mark both product and process will need to be developed, and advises against removing marks from open tasks, because marks motivate students [4]. A UQ case study in a TEQSA toolkit adds that criteria should speak to the learning a task is designed to show, and that descriptors aimed at spotting misconduct rarely work. Where AI use is not assessed, markers should grade the output only, and activities AI can automate, such as spelling, should weigh less unless core to the learning outcome [10].
Students are told in writing what is allowed. UNSW course outlines must state the allowed role of AI for each assessment [9]. At UQ, task instructions must state the tools students may use, and should ask for disclosure only where relevant [7]. Sydney requires students to acknowledge how they use AI [3].
Fairness and cost are part of the plan. TEQSA warns that falling back on invigilated, time-limited exams may reduce variety and authenticity and deepen existing inequities. It also notes the scheduling, supervision and documentation work that secure tasks in every subject can create [2]. Sydney asks that open tasks, wherever possible, be designed for all students from the start, to reduce individual adjustments [5].
5. What to take from this
For teachers and markers. The sources treat a written ban on AI in an open task as unenforceable [4, 11]. The work moves to the task and its rubric: what the task should show and how the process is marked [4, 10].
For heads of department and degree leaders. None of the three universities secures every task. Sydney and UNSW place secure points across the degree, and UQ sets a minimum share in each course [5, 8, 9]. Cost is an open issue: Sydney says adding secure tasks to more units would need considerable time, resources and extra workload [4].
For employers and training providers. At these universities, the secure points are where a student's own learning is confirmed [5, 8, 9]. At Sydney and UQ, open tasks allow or expect AI use [5, 8]. TEQSA's 2025 guide sets out three pathways for deciding where learning is assured [2].
6. Limits of this report
The cases come from one country and from official documents that describe rules and plans. This report's searches found no published evaluation of learning or misconduct under these models. UQ's rule and UNSW's secure-assessment clause start in 2027 and could change. Sydney's guidance pages have been edited since first publication; this report uses the 16 November 2026 versions. TEQSA's 2023 and 2026 papers call themselves advice outside its Guidance Notes [1, 13]. The 2025 paper on discursive changes was read only as an abstract [11].
References
[1] Lodge, J. M., Howard, S., Bearman, M., Dawson, P., & Associates (November 2023). Assessment reform for the age of artificial intelligence. Tertiary Education Quality and Standards Agency (TEQSA). https://www.teqsa.gov.au/sites/default/files/2023-09/assessment-reform-age-artificial-intelligence-discussion-paper.pdf
[2] Lodge, J. M., Bearman, M., Dawson, P., Gniel, H., Harper, R., Liu, D., McLean, J., Ucnik, L., & Associates (September 2025). Enacting assessment reform in a time of artificial intelligence. Tertiary Education Quality and Standards Agency (TEQSA), Australian Government. https://www.teqsa.gov.au/sites/default/files/2025-09/enacting-assessment-reform-in-a-time-of-artificial-intelligence.pdf
[3] The University of Sydney (27 November 2024). University of Sydney's AI assessment policy: protecting integrity and empowering students. https://www.sydney.edu.au/news-opinion/news/2024/11/27/university-of-sydney-ai-assessment-policy.html
[4] Bridgeman, A., & Liu, D. (2 July 2024). Frequently asked questions about the two-lane approach to assessment in the age of AI. Teaching@Sydney, The University of Sydney. https://educational-innovation.sydney.edu.au/teaching@sydney/?p=21264 (accessed 16 November 2026; page last modified 21 April 2026)
[5] Bridgeman, A., & Liu, D. (18 March 2025). The Sydney Assessment Framework. Teaching@Sydney, The University of Sydney. https://educational-innovation.sydney.edu.au/teaching@sydney/the-sydney-assessment-framework/ (accessed 16 November 2026; page last modified 26 July 2026)
[6] Bridgeman, A., & Liu, D. (2 October 2025). Navigating AI in Higher Education: tasks ahead for 2025 and 2026. Teaching@Sydney, The University of Sydney. https://educational-innovation.sydney.edu.au/teaching@sydney/?p=24402 (accessed 16 November 2026; page last modified 6 October 2026)
[7] The University of Queensland, Institute for Teaching and Learning Innovation. UQ’s rules for using AI. AI Teacher Hub. https://itali.uq.edu.au/node/11633 (accessed 16 November 2026)
[8] The University of Queensland. Assessment Procedure [version not yet in force; it takes effect from 1 January 2027]. Policy and Procedure Library. https://policies.uq.edu.au/document/view-current.php?id=184&version=10 (accessed 16 November 2026)
[9] UNSW Sydney (effective 1 June 2026). Assessment and Progression Policy [including the Assessment and Feedback Procedure, effective 25 September 2026]. https://www.unsw.edu.au/content/dam/pdfs/governance/policy/hub/assessmentandprogressionpolicy.pdf
[10] McGrath, D. (The University of Queensland). Principles for criteria and standards in assessment for gen AI use [Academic integrity toolkit case study]. Tertiary Education Quality and Standards Agency (TEQSA). https://www.teqsa.gov.au/guides-resources/protecting-academic-integrity/academic-integrity-toolkit/risks-academic-integrity-ai/principles-criteria-and-standards-assessment-gen-ai-use (accessed 16 November 2026)
[11] Corbin, T., Dawson, P., & Liu, D. (2025). Talk is cheap: why structural assessment changes are needed for a time of GenAI. Assessment & Evaluation in Higher Education, 50(7), 1087-1097. https://doi.org/10.1080/02602938.2025.2503964 (abstract read in the Macquarie University research portal record: https://researchers.mq.edu.au/en/publications/talk-is-cheap-why-structural-assessment-changes-are-needed-for-a-/)
[12] Tertiary Education Quality and Standards Agency. Gen AI strategies for Australian higher education: Emerging practice [announcement of the November 2024 toolkit, with the June 2024 request for information and its response rate]. https://www.teqsa.gov.au/about-us/news-and-events/latest-news/gen-ai-strategies-australian-higher-education-emerging-practice (accessed 16 November 2026). Toolkit: https://teqsa.gov.au/sites/default/files/2024-11/Gen-AI-strategies-emerging-practice-toolkit.pdf
[13] Lodge, J. M., de Barba, P., Ainscough, L., et al. (June 2026). Assuring quality learning in a gen AI-integrated future: The role of adaptive capabilities. Tertiary Education Quality and Standards Agency (TEQSA). https://www.teqsa.gov.au/sites/default/files/2026-06/assuring-quality-learning-in-a-gen-AI-integrated-future.pdf
[14] Liu, D., & Bridgeman, A. (12 July 2023). What to do about assessments if we can’t out-design or out-run AI? Teaching@Sydney, The University of Sydney. https://educational-innovation.sydney.edu.au/teaching@sydney/?p=18669 (accessed 16 November 2026; page updated May 2025)
[15] Steel, A. (12 July 2024). 2 lanes or 6 lanes? It depends on what you are driving: Use of AI in Assessment. UNSW Sydney, Education & Student Experience. https://education.unsw.edu.au/node/780
[16] UNSW Sydney (last updated 12 November 2025). Guidance on AI in assessment. UNSW Staff Teaching Gateway. https://www.teaching.unsw.edu.au/ai/ai-assessment-guidance
How to cite
IAPQB (2026). Secure and open assessment: how Australia's regulator and three universities are redesigning assessment for generative AI. IAPQB Report No. 11. International AI Practice Qualification Board. https://iapqb.org/research/011-secure-and-open-assessment.html
About this report
Prepared by IAPQB. Each fact comes from the sources listed, and each source was opened and checked on 16 November 2026; anything that could not be checked was left out. AI tools were used to search and to draft, and IAPQB checked every claim against its source and is responsible for the text. IAPQB runs an AI work qualification; this report does not promote it. Corrections: contact@iapqb.org or an issue at github.com/iapqb/iapqb.github.io. Licence: CC BY 4.0.
Archived with the series on Zenodo: doi.org/10.5281/zenodo.24248199 · Markdown sources: github.com/iapqb/research