Australia's seven state answers to AI cheating are now on one page

Digital Attitudes, 26 August 2026

Wednesday of Term 3 Week 7, and Tuesday's coverage put the Australian AI in schools story onto a single map for the first time this year. Two of today's three stories read the same data. The third one wrapped at ILUMINA Sydney the same afternoon.

The ABC finally maps the state by state AI enforcement patchwork, and NSW sits alone

ABC News published its cross jurisdictional roundup of Australian AI cheating enforcement on Tuesday 25 August, and reading the seven state and territory responses side by side is the first time the fragmentation has been laid out this cleanly in one place (ABC News on how education departments across the country are cracking down on students using AI to cheat). NSW is proposing to ban unsupervised take home assessment tasks for Years 11 and 12 and recorded more than 1,000 cheating cases in 2025. South Australia is the only jurisdiction that has built and shipped its own classroom AI (Microsoft co built EdChat) and is running the country's first AI Royal Commission from 1 October. Victoria has the VCAA guidance plus a school by school adjudication model plus the 90 minute primary screen cap landing Term 1 2027. Queensland requires academic integrity coursework via QCAA. Western Australia's Year 11 and 12 schools must implement senior secondary assessment policies naming AI. Tasmania and the Northern Territory sit on the federal framework and Tasmania blocks AI apps on departmental devices and networks. UTS Industry Professor Leslie Loble told the ABC that "AI is disrupting that process and keeping us much more in surface learning" and called for a national consensus.

Why this matters: this piece is the operational tell every Australian principal, deputy principal, and head of digital learning needs in front of them this week, because it makes the market segmentation official. A vendor selling one AI classroom product now has to pitch it to seven different regulatory perimeters, and a school buying one product now has to test it against the perimeter it actually sits inside. Three concrete Term 4 tasks for a Victorian or NSW head of teaching and learning. First, a one page comparison of the school's current AI use policy against the state framework the school actually sits under, because a NSW independent school still tied to NESA's cycle is running a different clock to a Victorian government primary. Second, an AI vendor register that names, per product, which state framework the vendor has documented compliance against, because a South Australian rollout with EdChat as the reference implementation is not the same procurement as a Tasmanian rollout with department network blocks against consumer AI. Third, a Term 4 assessment plan that names, per task, whether the task survives an unsupervised take home revocation in NSW, a 90 minute daily cap in Victoria, or a QCAA academic integrity clause in Queensland. The Loble line about surface learning and national consensus is the political ceiling for how far the sector can push a state fix without a federal answer. National Cabinet's AI standards process (Monday 17 August's brief) is where that answer arrives, and until it does the ABC map is the operational reality.

The UTS and UQ cognitive offloading paper is quietly becoming the citation the enforcement side leans on

Professor Leslie Loble AM (UTS) and Professor Jason Lodge (University of Queensland) published Artificial Intelligence, Cognitive Offloading and Implications for Education through the Australian Network for Quality Digital Education in March 2026 (UTS on experts warning unstructured AI use in schools risks 'cognitive atrophy'). The core argument is that unstructured AI use during formative learning short circuits the cognitive effort required for durable understanding, and the intervention the authors call for is fast adoption of national standards plus teacher capability to guide use. Loble's quote in Tuesday's ABC piece is not a stand alone comment. It is the two sentence version of the paper, and the paper is now cited on the record in most state and federal AI in schools submissions this year.

Why this matters: for a director of digital learning writing a Term 4 board paper, this is the citation the enforcement side of the debate is bringing to the argument, and if the paper is not in the school's AI policy references then the paper's argument is going to arrive via a parent or a journalist. Two operational reads. First, the paper does not argue for pulling generative AI out of the classroom, contra the Heads Up Alliance petition (Thursday's brief). It argues for structured use inside a national standards frame, which puts the intellectual centre of gravity closer to the ReadyTech benchmark position (Monday 18 August's brief on organisations without an AI use policy) and to Prue Car's NESA request (Wednesday's brief on the still unanswered letter) than to a full ban. Second, the paper is co authored by a Group of Eight professor at UQ and an industry professor at UTS who is also deputy chair of the country's national curriculum and assessment authority. That is exactly the pair of institutional signatures the National Cabinet AI standards process is likely to cite when it lands in early 2027, which means the argument is going to shape not just school policy but the standards themselves. The Term 3 action is to read the paper this week rather than to be quoted its findings back to you in Term 4.

CanvasCon ANZ closed on Tuesday with the AI capability and neuroscience headliners, not the vendor

Instructure's CanvasCon ANZ 2026 wrapped at ILUMINA on Elizabeth Street Sydney Tuesday afternoon after the two day run that opened Monday, and the two keynote speakers on the podium were UNSW Sydney ARC Laureate Fellow and Scientia Professor of AI Toby Walsh and University of Tasmania senior lecturer and neuroscientist Dr Lila Landowski, alongside Instructure global chief executive Rachel Orston (iTWire on Toby Walsh and Lila Landowski headlining CanvasCon ANZ 2026 in Sydney). The theme on the stage banner read Shaping Readiness, Together, and the roster of participating institutions ran to UNSW Sydney, the University of Melbourne, the University of Sydney, RMIT, Griffith, Adelaide University, the University of Newcastle, Swinburne and the University of Auckland, alongside schools, VET providers and other education organisations.

Why this matters: yesterday's brief covered the Instructure survey findings released to coincide with the conference. Tuesday's actual stage told a slightly different story: the vendor put a Scientia Professor of AI and a neuroscientist in front of the room, not a Chief Product Officer with a roadmap slide. For a director of digital learning who did not attend, that programming choice is the tell. Two reads. First, Instructure walked into its first ANZ conference since the April to May Canvas breach with a public credibility problem, and putting Walsh and Landowski on the podium is the same play the sector's Group of Eight vice chancellors reach for when the tenant conversation gets uncomfortable: put an independent academic voice on the record before the product roadmap. Second, the participating institutions list is a Canvas customer roll for Australian tertiary that reads as a national reference customer set, and the visible absence of a live product announcement means the follow up account meetings that start this Friday are the actual news moment, not the keynote. The LinkedIn traffic from the two days will decide whether Australian Canvas renewal conversations this term hold as normal or need a security governance workstream bolted on before Term 4. For a non attending head of learning technology, the Term 3 action is a five minute reach out to a colleague who was in the room to ask what Instructure signed to on contract language, because that is the thing the podium did not answer.

Three signals, one Wednesday, and the theme is that Australian AI in education is running on seven state maps, one national argument that just got its intellectual anchor, and a vendor conference that programmed around its own hardest question. More on the operational landscape at digitalattitudes.com.au.