Joint AI committee launches as universities face smart glasses gap

Digital Attitudes, 21 August 2026

Friday of Term 3 Week 6, and the last 48 hours have added a Commonwealth committee, a wearable device gap and a cyber vendor's blunt warning to the Australian AI in education operational file.

Government backs the Coalition's Joint Select Committee on AI

Shadow Industry Minister Andrew Hastie, Shadow Home Affairs Minister Jonathon Duniam and Shadow Digital Economy Minister Aaron Violi confirmed on Wednesday 20 August 2026 that the government will back the Coalition's move to establish a Joint Select Committee on Artificial Intelligence (Liberal Party release on the Joint Select Committee on AI). The stated scope is national security, productivity, competitiveness and living standards, with the adequacy of existing laws, data sovereignty and the impact on Australians named as focus areas. Hastie's framing on the record: "artificial intelligence poses moral, strategic and economic challenges to Australia and our people." Independent coverage from MLex confirms the government endorsement of the opposition proposal (MLex on the new parliamentary AI committee).

Why this matters: this is the third Australian AI oversight venue opened in a month, sitting alongside the South Australian Royal Commission (Friday 15 August's brief) which formally launches 1 October, and the National Cabinet AI standards process (Monday 17 August's brief) still expected to legislate in early 2027. For a school principal, university CIO or state department director of digital learning, the practical implication is that any AI product bought this term is now going to have to survive three sets of public evidence within twelve months. Two operational tasks this week. First, the vendor evaluation template needs a "will the vendor testify on the record if asked" question added to it, because a Joint Select Committee has the standing to call submissions from tenants as well as vendors. Second, the internal AI register the ReadyTech benchmark (Monday's brief) said two thirds of Australian education organisations do not have needs to be written this term, not next, because a bipartisan committee producing a report in the first half of 2027 will reference organisational registers when it asks whether the sector could account for what it was running.

Smart glasses expose a blind spot in university policy

Janine Arantes, Bec Marland and Andrew Welsman at Victoria University published a piece in Future Campus this morning arguing Australian universities have written smart glasses policy for the exam room and forgotten the rest of campus (Future Campus on the smart glasses university policy gap). ANU prohibits smart glasses as electronic devices in exams, RMIT bans them from exam venues, Adelaide University requires removal. UTS researchers are building assistive spatial audio applications for blind and low vision users on the same hardware, Monash is researching cerebral visual impairment support, and the University of Sydney is studying gendered harassment risks including covert recording. The authors' argument is that fragmented accessibility, privacy and conduct policies operating separately leave staff without operational clarity on whether a lecturer can ask a student about recording status, how the rules differ in a classroom versus a laboratory versus a clinical placement, and what a staff member does when a concern is raised.

Why this matters: this sits directly on top of Victoria's Education and Training Reform Amendment Act 2026 (Tuesday's brief) which extended the personal electronic devices ban to non-government schools and named wearables and personal audio devices explicitly, with a 1 July 2027 deadline. The K to 12 sector has an eleven month operational clock and an Act that names the device class. The Australian higher education sector has neither, and the Future Campus piece names six sandstones handling the same device six different ways. Three concrete tasks for a pro vice chancellor academic. First, the smart glasses policy has to be written as one document by the accessibility, privacy and conduct teams together, because the current fragmentation is the piece the authors flag as most dangerous. Second, the operational guidance for teaching staff needs to be plain English and one page: what to ask, when to escalate, when not to assume misuse. Third, the eSafety Commissioner's 28 July advisory on AI generated imagery in schools (covered in prior briefs) applies here too, because a covertly recorded lecture becomes a deepfake without the recorder needing to leave the room.

AI phishing has arrived at Australian universities, and the training deck is out of date

Matt Caffrey published "Phishing goes next level with AI" in Campus Review on Tuesday 19 August 2026, and the core argument is that the "look for bad grammar and odd formatting" cybersecurity training slide Australian universities have been running for a decade is now obsolete (Campus Review on AI generated phishing at universities). Caffrey's line on the record: "that advice is rapidly becoming obsolete." Generative models write clean, contextual, targeted email at scale, and the volume plus the quality plus the personalisation is producing convincing lures aimed at institutional staff, not the retail consumer the training was designed for. Caffrey's framing pushes the response up the org chart: cyber defence has to be a leadership priority rather than a technical implementation, because the training and the technical controls both need refreshing at the same time.

Why this matters: three Australian tertiary and TAFE data incidents in the last four months (the Canvas breach that swept up the Australian Computer Society and multiple universities, the Reynella East College Adelaide dump of over 600 gigabytes in June, and UWA's June exposure of student information system credentials) all began or could have begun with a plausible phishing email into a staff or student account. For a university CIO or a director of digital learning, the practical Term 3 action is a one page audit of the current phishing training. Is it still teaching "spot the typo" as the primary tell? Is the simulated phishing email being generated by a template from 2020 or by a current generation model? Is a staff member reporting a well written and contextually accurate lure being told they were wrong because the grammar was fine? If any of those answers is uncomfortable, the training is now part of the risk, not the mitigation. This also sits directly under the OAIC Children's Online Privacy Code deadline of 10 December 2026 (less than four months out) which will be read by regulators against the organisation's actual staff training when the first breach after 10 December lands.

Three signals, one week, and the theme is that Australian AI in education is being audited from three different directions at once: by a Commonwealth committee, by a policy vacuum on a device class that is already in the room, and by a threat model the vendor training material has not caught up with. More on the operational landscape at digitalattitudes.com.au.