Gemini defaults on in Australian Google Classrooms

Digital Attitudes, 24 August 2026

Monday of Term 3 Week 7, and three of the last week's moves put Google on the front page of the Australian ed-tech operational file: a Classroom default that opts students in without an admin decision, a free consumer AI tier for university students, and a locally built AI stack from a Brisbane K-12 principal running against the grain of every other story in this brief.

Gemini in Google Classroom is now on by default, mobile rollout began last Monday

Google's Workspace Updates blog confirmed on 4 August that Gemini in Google Classroom is expanding to students of all ages, with web rollout from 10 August and mobile rollout from Monday 17 August 2026 (Google Workspace Updates on the Gemini in Classroom rollout). The feature is on by default for teachers and students of any age via the Gemini in Classroom control, and administrators have to actively disable it by organisational unit to block younger year levels. Contextualised starter prompts pull assignment titles, instructions and curriculum materials directly from Classroom, and the release ships flashcards, practice quizzes, a "Learn with Gemini" tile on Due Soon items, and Gemini Notebook integration.

Why this matters: for any Australian school running Google Workspace for Education (the default stack in most K to 12 systems outside NSW), Gemini is now live for students on mobile as of last Monday unless the domain admin has opted the organisational unit out. That is a de facto AI in classroom deployment that most principals did not sign off on and most parents were not notified about. Two Term 3 tasks for a Victorian, Queensland or WA head of digital learning. First, check with the domain admin whether the Gemini in Classroom control is on for the primary organisational unit, because the answer determines what the current Term 3 assessment schedule is actually running under. Second, the AI use policy the ReadyTech benchmark (last Monday's brief) said two thirds of Australian education organisations do not have needs to name Google Classroom explicitly, because the personal versus approved tool framing that Steiner (last Tuesday's brief) and the Heads Up petition (last Thursday's brief) have both leaned on collapses when the approved LMS ships its own AI by default.

Google AI Plus goes free for a year for Australian university students

Google's back to school announcement on 20 August 2026 extends its Google AI Plus tier to eligible college students in more than 140 markets, Australia included, for a full year at no charge, with 400 GB of storage and higher usage limits on Gemini bundled in (Google's blog on the Google AI student offer, iTWire on the Australian availability). Sign-up runs to 31 December 2026. The announcement pairs with a redesigned Gemini student hub featuring diagnostic quizzes that produce tailored study plans, and Deep Research reports inside Gemini Live.

Why this matters: this puts a free consumer grade AI tier in the pocket of every Australian undergraduate before the sanctioned university stack has finished its rollout. UNSW is on ChatGPT Edu (10,000 licence staff deployment since September 2025), La Trobe is stepping through 5,000 licences in 2026 to 40,000 by 2027 alongside its AI MBA, Monash is on Microsoft Copilot for students, and USYD's Green Paper argues for assessment redesign over a compulsory module. None of those procurements can compete with a free tier that a student activates in five minutes on a personal Google account. The operational tell for a pro vice chancellor teaching and learning is that any 2027 licence conversation now has to answer why the university tier is worth the money when the same student has a bundled Gemini Advanced sitting on their phone. The safer read is that the argument is now about data governance, integration with the LMS and academic integrity workflows, not about model access.

Islamic College of Brisbane builds Aladin, its own predictive analytics and AI stack

Islamic College of Brisbane chief executive Ali Kadri confirmed in coverage published on 17 August 2026 that the Karawatha school is developing its own in-house AI assistant, Aladin, a play on the Iron Man tool Jarvis, designed to generate classroom learning materials and to flag students who need intervention on the basis of historical grades and attendance (Oz Arab Media on alternative schools weighing AI's role in learning). When the model projects a slip, it fires an alert to a staff member who then decides on the intervention. The story sits inside the same alternative schools cycle that carried the Steiner Education Australia opt out covered in last Tuesday's brief.

Why this matters: this is the rare Australian K to 12 case study of a school building its own predictive analytics and generative AI stack in house rather than buying Corella (Queensland), NSWEduChat, EdChat (South Australia), Cogniti (USYD Azure Marketplace listing) or a Microsoft Copilot tenant. For a principal or director of digital learning weighing a Term 4 vendor conversation, Aladin is the Australian counterexample to the vendor pitch that only a state department or a Group of Eight has the capacity to build this in house. It also puts the predictive analytics question directly in front of school governance, because a model that flags a Year 8 student as at risk on the basis of attendance and past grades is a decision surface with well documented failure modes, and Kadri's build costs an ethics conversation that a Corella tenant does not. The Term 3 lesson is that the barrier to a school building its own tool has fallen enough that a single school CEO can ship one, which changes what a principal has to be able to answer when the school council asks why the school is on a vendor rather than a local build.

Oral vivas as an AI defence carry a real equity cost

University of Southern Queensland lecturer Nicole Brownlie published a piece in The Conversation on 17 August 2026 arguing that the current move back to oral vivas as a defence against generative AI in assessment has an evidence base that is thinner than the sector's current enthusiasm implies (The Conversation on the return of oral exams). Her cited studies include a German paper reporting higher student anxiety in oral compared with written exams, and a Norwegian secondary school study documenting extreme variability: presentations ran 5 to 16 minutes, follow up discussions 7 to 23 minutes, and examiners asked between 10 and nearly 50 questions of different students in the same cohort. Brownlie's argument lands as Melbourne Year 12 assessments in the past fortnight have caught students using AI on unsupervised tasks.

Why this matters: this piece is the analysis anchor for the NESA moratorium debate still running on the Term 4 clock (last Wednesday's brief on NESA's silence, Saturday's brief on Dr Nira Rahman's Melbourne piece). For a NSW head of teaching and learning drafting a Term 4 assessment plan on the assumption that oral vivas are the safe fallback, the equity read is the hard one. A 5 to 16 minute range and a 10 to 50 question range for the same cohort is not a controlled instrument, it is a moderation nightmare, and the German data on anxiety puts the students who most need a fair assessment surface at the wrong end of the distribution. The practical Term 4 action is that any assessment plan pivoting to oral vivas needs a written examiner script and a stopwatch protocol before it goes to parents in Week 8, or the school is trading an AI integrity problem for a moderation and equity problem it will spend Term 1 2027 defending.

Four moves, one week, and the throughline is that Google has quietly become the largest AI in Australian education vendor by default, one Brisbane principal has shown that a local build is now feasible, and the assessment reform story is running out of easy alternatives. More on the operational landscape at digitalattitudes.com.au.