Australia’s higher-education regulator has told universities to stop leaning on artificial-intelligence detection software to catch cheats, and to redesign their assessments instead. In March 2026 the Tertiary Education Quality and Standards Agency (TEQSA) published a revised Academic Integrity Toolkit with two new sections, one on assessment design and security, the other on the risks generative AI poses to academic integrity.
The message beneath the update is blunt. AI detectors cannot guarantee integrity, and a flag from one should not be treated as proof a student cheated. That reframes the sector’s response to generative AI from surveillance to design, and it lands as a national policy signal for all 43 registered universities, not a suggestion.
Detectors on the way out
The regulator’s caution reflects what is already happening on campuses. Curtin University in Perth switched off Turnitin’s AI writing detection from 1 January 2026. In its own notice to students, Curtin said only the generative-AI detection feature would be disabled, while ordinary text-matching originality checks would remain active. The university framed the move as being about “fostering trust and clarity within a modern academic culture” and keeping assessments “secure, fair, relevant and future-ready”.
Curtin is not alone. According to reporting collated by AcademicJobs, more than a dozen Australian universities have wrestled with AI-detection false positives, with the University of Queensland having earlier turned its tool off and the Queensland University of Technology among those reporting cases where students were wrongly flagged. The same account documents the human cost: transcripts marked “results withheld” for months, delayed graduate registration for nursing and law students, and students asked to hand over browser histories and drafts to prove their own innocence.
Those complaints track a wider evidence problem. Detection vendors advertise very low false-positive rates, but independent studies have reported far higher figures, and researchers have repeatedly found the tools flag writing by non-native English speakers more often. A 2026 paper in the Journal of Higher Education Policy and Management captured the bind in its title, “Heads we win, tails you lose”, arguing that detectors put students in a position where the technology, not the evidence, decides guilt.
What TEQSA is asking for instead
Rather than a better detector, the regulator wants better assessment. TEQSA’s toolkit and its companion guidance describe several models institutions are testing, summarised by learning-platform vendor Instructure as three pathways: redesigning assessment across a whole degree, embedding at least one secure task in every subject, or a hybrid of program-level oversight and unit-level checks. The common thread is that at least some assessment happens in conditions the institution can vouch for, through in-person tasks, oral defences and staged, process-based work that is harder to automate.
The toolkit also shifts the burden of proof in a practical direction. TEQSA encourages students to document how they used generative AI, including the prompts they entered and how they folded any output into their own work, effectively asking them to show their working. That advice does two things at once. It gives honest students a way to demonstrate authorship without relying on a detector’s verdict, and it turns AI use into something to be declared and assessed rather than hidden.
TEQSA extended the argument on 24 June 2026 with a further resource, Assuring quality learning in a gen AI-integrated future: The role of adaptive capabilities. The paper argues that in a world where generative AI is embedded in study and work, the point of assessment is to develop and verify capabilities that AI cannot supply on a student’s behalf, such as evaluative judgement, critical thinking and ethical reasoning.
Why it matters for Australia
This is the national regulator, not an individual campus, setting the direction, and that carries weight. TEQSA administers the Higher Education Standards Framework, so guidance that reframes integrity as an assessment-design problem effectively sets the compliance benchmark every provider will be measured against. Universities that keep treating a detector flag as evidence of misconduct now sit further from the regulator’s stated expectations.
The stakes are practical. Australian higher education leans heavily on international students, many of whom write in English as an additional language and are precisely the group most exposed to false positives. A detection-led approach risks penalising them for their writing style, with consequences that reach into visas, registration and employment. Steering the sector away from that reduces a real equity and reputational risk for a system that trades on the fairness of its qualifications.
It also resets what universities are buying and building. Money and effort spent chasing an unreliable detector can instead go into secure exams, oral assessment, authentic tasks and the marking capacity they require. That is a larger, slower investment than switching on a software feature, and it is where the harder work of the next few years will sit.
The unresolved question is scale. Redesigning assessment across crowded programs, with large cohorts and stretched teaching staff, is expensive and time-consuming, and TEQSA’s guidance sets a direction without funding the transition. Expect the next phase to be less about whether detectors are trusted, and more about which universities can afford to rebuild assessment quickly, and how the regulator judges those that cannot.
Sources: TEQSA — Revised Academic Integrity Toolkit now available; TEQSA — Academic Integrity Toolkit; TEQSA — Assuring quality learning in a gen AI-integrated future; Curtin University — Update on Turnitin AI-detection tool; Instructure — Three pathways to assessment reform; AcademicJobs — AI cheating false accusations at Australian universities; Journal of Higher Education Policy and Management — Heads we win, tails you lose.
