NAPLAN gap widens as Gemini opens to every student

Digital Attitudes, 5 September 2026

Saturday, and this week's file is three moves at three different altitudes. The national assessment reporting stack (NAPLAN), the classroom AI vendor stack (Gemini), and the individual staff member reaching for the same tools their employer is quietly buying (Macquarie). All three are the same story from different chairs.

NAPLAN 2026 equity gap widens on almost every axis, and Year 5 numeracy is the only bright spot

The 2026 NAPLAN analysis from Mitchell Institute senior research fellow Andres Molina at Victoria University, surfaced by The Educator K/12 on Tuesday 2 September 2026, tracks the gap between students whose parents completed a bachelor's degree and those whose parents did not complete school, and the gap grew across four of the five reading and numeracy year levels compared with 2023 (The Educator K/12 on the widening NAPLAN gap, EduResearch Matters on what the 2026 results really tell us). Year 3 reading went from 1 year 10 months behind to 2 years 2 months. Year 7 reading from 4 years 11 months to 5 years 6 months. Year 9 reading from 6 years to 6 years 8 months. Year 7 numeracy from 5 years 3 months to 5 years 11 months. Year 9 numeracy from 6 years 8 months to 7 years 9 months. Year 5 numeracy is the sole gap that narrowed, from 2 years 10 months to 2 years 5 months. Participation is now a story in itself: Year 7 numeracy participation was 91.3 per cent nationally, 76.6 per cent for Indigenous students, 79.7 per cent in remote schools and 62.9 per cent in very remote schools. In major cities, 70 per cent of students met exceeding or strong standards. In very remote schools, 23 per cent did. ACARA chief executive Stephen Gniel on the record: "gaps remain the system's toughest challenge".

Why this matters: for a school business manager, a head of teaching and learning, and a principal building a 2027 assessment and improvement plan, the widening happened during the same three years the sector spent on generative AI framework work and cyber breach response, and the equity dial moved in the wrong direction anyway. Two operational reads. First, the data warehouse question. If the school analytics stack still stops at Scout, PAT, and the school's own dashboards, and does not tie the NAPLAN result to the Index of Community Socio-Educational Advantage quartile and the actual attendance and English as an Additional Language flag for the same student, the school is running the same rear view mirror that produced this national picture. The Mitchell Institute analysis works because the data is disaggregated, and the school-level version of that analysis is the piece a 2027 improvement plan sits on. Second, the participation number. A Year 9 numeracy cohort where the very remote participation rate is 62.9 per cent is a cohort where the reported average is not the whole picture, and any procurement conversation about adaptive learning platforms or AI tutoring products in 2027 needs to answer for who the tool moves and who it does not reach. The 23 per cent very remote figure does not get fixed by a chatbot subscription. It gets fixed by the boring parts of school infrastructure that vendors rarely lead with.

Google flips Gemini in Classroom on by default for every K-12 student, and the admin OU is now homework

Google's Workspace Updates blog confirmed on Tuesday 4 August 2026 that Gemini in Google Classroom is now available to K-12 and higher education students of all ages, with the web rollout from Monday 10 August 2026 and mobile from Monday 17 August 2026, and the "Gemini in Classroom" control is enabled by default across Education Fundamentals, Standard and Plus editions (Google Workspace Updates on the K-12 all ages Gemini expansion, Education Week on schools caught flat footed by the Google move). To restrict access for students under 18, an administrator has to create a separate organisational unit and turn off Gemini for that OU. Robbie Torney at the Youth AI Safety Institute framed the shift on the record: the rollout "defaults to adoption rather than permission". Michael Martin, principal of Plymouth High School in Ohio, told Education Week his school turned Gemini off, because the teacher training was not in place, and he learned about the change from a news story rather than from Google. Gemini is now the only major AI chatbot with default availability to K-12 students of every age. Microsoft Copilot still requires admin activation and a minimum age of 13.

Why this matters: for an Australian government, Catholic or Independent school running a Google Workspace for Education tenant, the operational default flipped last month, and the assumption that a Term 4 AI use policy governs what students can reach in the classroom is now only as strong as the OU structure in Google Admin. Two operational reads. First, the audit question every IT lead should have already answered but many have not. Is Gemini in Classroom on for students in the primary campus tenant, on for Year 7 to 9 who are under 13, and on for the senior secondary cohort where policy might allow it. If the answer is "we did not touch it", the answer is "on for everyone". The OU exercise is not a Term 4 improvement project. It is a compliance task under the school's existing Acceptable Use Policy and the Australian Framework for Generative AI in Schools that has to be logged as done. Second, the vendor procurement read. Corella, NSWEduChat, EdChat and the various sector-specific AI tools have spent 18 months arguing their case on data residency, curriculum alignment and safety guardrails. Google's move re-anchors the comparison, because a school with a live Workspace tenant now has a default-on general purpose chatbot from a hyperscaler in the same classroom the state tool sits in. The 2027 procurement conversation is no longer "which single AI tool do we adopt". It is "which tool do we deliberately point students at, given the general purpose one is already on the desk".

A Macquarie IT lecturer wins his Fair Work casual conversion case using ChatGPT, and the same faculty runs the Virtual Peer chatbot

Greg Baker, an IT lecturer in Macquarie University's School of Computing and a casual academic since 2023, self-represented in a Fair Work Commission casual conversion case and won, with the Commission finding his work constituted "recurring fixed term periods of part time employment, rather than genuine casual employment", and Baker credited a $280 USD per month ChatGPT Pro subscription with doing "very in depth work, reading through hundreds of Fair Work rulings and finding neat analogies" (ACS Information Age on the Macquarie IT lecturer winning Fair Work using AI). Baker on the record: "Nobody was more surprised than me when ChatGPT won", and separately, "Here's AI, it can bring about access to justice, it can change the balance of power". The Fair Work Commission itself has publicly warned about "AI slop" from inaccurate AI generated submissions, so the finding sits inside a wider tribunal conversation about whether AI assisted self representation is a democratising force or a workload risk.

Why this matters: for a pro vice chancellor academic at any Australian university, and for every Group of Eight and ATN human resources office reading the NTEU picket at USyd and the Guardian coverage of Macquarie's Virtual Peer psychology tutorials as background (Wednesday and Thursday's briefs), this is the same story from a third chair. The tool the university is buying to reduce staff hours in tutorials is also the tool a casual employee at the same institution just used to prove he was not a casual. Two operational reads. First, the enterprise agreement conversation the NTEU walked USyd staff off the job over on 2 September is not only about jobs the AI takes. It is also about the standing an individual staff member has when they use the same tool to file, and win. Any AI clause bargained into a 2027 agreement now has to address both directions of use, because Baker's win is a live example. Second, the procurement paper for any 2027 AI marking, AI tutoring or AI feedback product should include a workplace impact section that names both. What does the vendor deployment reduce, and what does the same tool, in the hands of the staff member whose hours it reduces, enable. If the vendor's slide has an answer for one but not the other, the procurement paper is incomplete. The Baker case does not decide the enterprise agreement conversation, but it does tell the finance office that a workplace AI conversation confined to the "how do we deploy" side of the ledger will be answered by staff on the "how do we use" side, whether the university plans for it or not.

Three moves, one Saturday. The national assessment picture confirms the equity gap grew during the same three years the sector spent on frameworks. The classroom vendor default just flipped, and the Term 4 AI Acceptable Use policy is only as strong as the school's Google OU structure. And a casual IT lecturer at Macquarie used the same AI stack his employer is buying to win a Fair Work case. Prior editions and the fuller operational read live at digitalattitudes.com.au.