AI in Schools
Mitchell Institute report finds five-year NAPLAN gaps
The Thursday read is one report that turns 17 years of NAPLAN into a structural argument, one US Committee briefing deadline that arrives at close of business today, one conference floor conversation worth carrying into the next procurement cycle, and one OAIC clock with fifteen days to run.
Mitchell's 17 years of NAPLAN data lands a structural argument
The Mitchell Institute at Victoria University released Years apart: Australia's growing educational inequality on Monday 18 May, drawing on NAPLAN data from 2008 to 2022. The analysis uses an Equivalent Years of Learning approach to convert raw scores into years and months relative to the Australian average, and finds the reading gap between Year 5 students whose parents hold a bachelor's degree and those whose parents did not complete school has grown from two years and three months in 2008 to three years and two months in 2022. By Year 9, the average reading gap between the most and least advantaged students sits at four years and three months, and on some sub-measures the gap is wider. Lead author Dr Andres Molina says "the enduring nature of these learning gaps suggests the issue is structural" and is "driven by policies at the system level".
Why this matters: For school IT leaders and data teams running performance dashboards against the National Assessment Program, this is the first cross-state analysis in three years that uses a method other than scaled scores, and the Equivalent Years of Learning frame is the one that will start appearing in board papers and council briefings through term three. For the Commonwealth, the report is the public-domain case for the $18.2 million NAPLAN platform allocation announced in last week's Budget to actually do something, namely, hold the test stable enough to give the trend data a reliable next decade. For analytics product teams selling adaptive content into Australian schools, the procurement read is sharper: the report names parental education as the dominant predictor across the period, and a tool that cannot demonstrate it closes that gap rather than widens it will not survive a 2027 tender against the new ATEC equity-evidence bar. Years apart on the Mitchell Institute site, and Victoria University's release on the widening advantage gap.
The Canvas briefing deadline arrives at close of business today
US House Homeland Security Committee chair Andrew Garbarino asked Instructure CEO Steve Daly on 11 May to brief the committee by close of business today on both Canvas intrusions, the data accessed, containment, notification, and Instructure's coordination with CISA and federal law enforcement. As of Tuesday Instructure had not publicly confirmed whether Daly or a senior representative would appear, and the format Garbarino requested is a closed-door briefing rather than open testimony. The deadline lands ten days after Instructure's 12 May statement that ShinyHunters had returned and destroyed the 3.65 terabyte dataset, and the committee's invitation letter is explicit that it sees "systemic vulnerabilities" worth examining.
Why this matters: For Australian vice-chancellors and chief operating officers, the read this morning is which procurement clauses the briefing transcripts (or the absence of them) will end up changing. A closed-door brief is not a public hearing, but the documents the committee tables and the line of questioning that follows shape what every other ed-tech vendor will have to answer when an Australian university council asks the same question in the September renewal window. For Australian school IT leaders, the question this morning is whether the institution's Notifiable Data Breaches assessment file is ready in case Friday or next week produces a second wave of disclosure obligations. The slow part of the breach story is the regulatory tail, and the tail just got longer. SC Media on the House committee calling Instructure to testify, and Dark Reading on the committee's framing of the briefing.
Tuesday's AI in Education Sydney, two days on
Informa's Artificial Intelligence in Education NSW conference ran all day at Swissôtel Sydney on Tuesday 19 May, with NSW Department of Education's Jason Miezis, ACARA's Melanie Hughes, and Macquarie's Professor Matt Bower in the three keynote slots. Two days on, the floor conversation worth carrying into the term three procurement cycle was not about which tool to buy but about who owns the implementation risk inside the school. The recurring question across the breakout streams was who in the school is accountable if an AI-recommended differentiation pathway turns out to widen the gap rather than close it, which is the same accountability question the Mitchell Institute report puts to the system.
Why this matters: For independent and Catholic school heads, the post-conference question to take to the executive is the accountability one, which named role inside the school owns the outcome of an AI-mediated decision (a recommended pathway, an automated comment, a flagged student) and how that accountability is documented in the staff handbook. For state-system procurement teams, the Tuesday signal was that NSW EduChat, the South Australian EdChat, and the WA ClassmAIte models are now read alongside one another as the three Australian state-built reference architectures, and the 2027 cross-jurisdictional procurement question has moved from "build or buy" to "which state-built reference, and on what licensing terms". For the vendor side, the floor read was that tools that cannot trace a recommendation back to a defensible evidence base will be marked down in the next two procurement cycles. The Bower keynote on anticipatory governance was the academic spine; the Miezis and Hughes sessions were the operational ones. The Informa Australia conference page for the NSW event.
Fifteen days on the OAIC privacy code, and the schools sector is still under-represented
The OAIC's draft Children's Online Privacy Code closes for industry, civil society and academic submissions at close of business Friday 5 June, fifteen days from this morning. The published submission register so far is heavy on the ed-tech vendor side (Pixevety, Snap, the platform players) and on the civil-society side, and lighter on the school-sector and university-sector side. The Code covers any service likely to be accessed by children, which means most of the ed-tech stack an Australian school operates today, and registers by 10 December 2026. The OAIC has been clear that the regulator expects the Code to land with a built-in best-interest assessment, restrictions on profiling, and a statutory deletion right that ed-tech platforms have to honour by default.
Why this matters: For school business managers and university registrars, the under-representation in the submission register is the operational risk. A code that lands in December without strong school-sector input will be a code the schools side then has to comply with under vendor-led interpretations, and the contract re-papering work in the first quarter of 2027 will be harder than it needs to be. The work this fortnight is a short, specific position from each school network or diocese, naming the operational consent burden a multi-platform parent already carries and the deletion-rights gap on platforms that hold longitudinal student data. The OAIC has been clear that the regulator wants to see the schools sector on the file, not just the lawyers and the platforms. The OAIC's Children's Online Privacy Code register page, and Gilbert + Tobin's read of what the exposure draft means for service providers.
The thread today is the data that sets the question (Mitchell's NAPLAN read), the vendor accountability that is being tested in real time (Canvas at Congress), the conference floor that has just landed on the same accountability question one level down (the Tuesday Sydney conversation), and the regulator window that decides who answers the question in writing (the OAIC code). More on the operational landscape at digitalattitudes.com.au.