USyd rally rejects OpenAI as privacy tranche 2 closes

Digital Attitudes, 17 September 2026

Thursday morning, and the week's read is a governance one: a Sydney signature that skipped consultation, a Canberra consent bill that closes at the end of the week, a Mathspace patch clock the ASD says was thirty times too slow, and a PISA number that says daily AI use in schoolwork costs students what once-a-week use does not.

USyd students snap rally against the OpenAI deal, and the NTEU says it was never consulted

Two days after the University of Sydney announced its multi-year ChatGPT Edu partnership with OpenAI, students held a snap rally on Eastern Avenue on Wednesday 16 September, chaired by SRC President Grace Street, with NTEU professional staff and Greens senator Mehreen Faruqi on the microphone (Honi Soit on the Wednesday rally and the NTEU's line that it was not consulted on the procurement). Faruqi's line on the record: "Major changes on campus should be decided by the people who actually make up universities: the students and staff. They should not be dictated by managers cutting closed-door deals with mega corporations". USyd's own coverage confirms the partnership was only "noted" at the 8 September Academic Board meeting, with Vice-President Nicole Gower attributing the rollout to a Responsible AI Framework launched on 10 June 2026 (The Educator on Mark Scott's announcement and the Nicole Gower governance framing).

Why this matters: for an Australian DVC, provost, or head of teaching and learning writing a 2027 AI framework this term, USyd is now the reference sequence. A signature first (Sunday), an Academic Board noting rather than approving (a week before), a Responsible AI Framework that binds implementation but did not shape the buy (three months earlier), and a staff and student response that arrives in the same week as the press release. Two operational reads. First, the governance one. Any 2027 AI framework whose only trigger is post-signature "implementation" guardrails will be tested against the same protest and press cycle USyd is running through, and the enterprise-bargaining rounds already open in the sector give the NTEU the mechanism to demand written consultation. The specific paragraph the September or October council paper needs is a named consultation trigger before signature, not after, with the Academic Board and the union both on the list. Second, the vendor one. OpenAI now has three publicly announced Australian ChatGPT Edu deals in nine months (La Trobe December 2025, UNSW Sydney on 10 September, USyd on 14 September), each in the 40,000 to 80,000 licence range, and Anthropic's Claude for Teachers enterprise offer has been free to K to 12 systems since 28 August with a June 2027 sign-up window. A university leadership team writing its own OpenAI, Anthropic or Google terms this term should assume the fourth Australian deal is being scoped this week, and the room to negotiate a bespoke consultation clause shrinks with each announcement.

Australia's privacy tranche 2 exposure draft closes on Friday, and the "fair and reasonable" test is on the school procurement paper

The Attorney-General's Department opened the exposure draft of the Privacy Amendment (Personal Data Protection) Bill 2026 on 31 August, and public submissions close at 5pm Friday 18 September (Clayton Utz on the tranche 2 proposals and the seven "fair and reasonable" factors, IAPP on the second-wave reforms and the 72 hour breach clock). The bill folds the 13 Australian Privacy Principles into a unified "fair and reasonable" test whose legislated factors include reasonable expectations, transparency, data minimisation, genuine choice, proportionality, and the best interests of the child. It lifts breach notification to 72 hours from the current 30 day assessment window, adds a positive obligation on entities to take reasonable steps to contain a breach, and creates a right to erasure for entities above the $500 million revenue or 2.5 million monthly user thresholds. It leaves the Children's Online Privacy Code register-by-December work on the separate tranche 1 track.

Why this matters: for a school business manager, director of digital learning, or university privacy officer, three practical items land on the September or October council paper. First, the "fair and reasonable" test is the vendor-contract paper. Any current or 2027-renewal ed-tech contract that treats a broad consent block as a licence for downstream analytics or third-party sharing has to answer the seven factors on the front page of the record, with "best interests of the child" the specific K to 12 test. Compass, Sentral, Schoolbox, SEQTA, Canvas, Mathletics, Reading Eggs, Studyladder, IXL, Essential Assessment, Corella and NSWEduChat all sit on this list. Second, the 72 hour breach clock is the Mathspace lesson made law: the Term 4 breach response playbook needs the 72 hour timer built into the incident response, not the 30 day assessment window that the OAIC scheme currently allows. Third, if the school system exceeds the $500 million or 2.5 million thresholds (state department platforms, national LMS providers, some university tenants), the right to erasure becomes an operational service the vendor has to build against a legislated request path. Submissions close 5pm Friday. Any leader in the sector who has a specific pattern the bill needs to address (the bundled "internal platform plus public social media" consent that has been the WA schools flashpoint is one) has forty-eight hours to say so on the record.

The Educator puts the Mathspace patch clock against the ASD Essential Eight and gets a thirty-times gap

The Educator K/12 published a follow-up on 14 September framing the Mathspace breach against the Australian Signals Directorate's Essential Eight, which recommends internet-facing systems be patched within 48 hours of a critical vulnerability being identified (The Educator on the 23 day patch gap in school systems and the ASD standard it fails). The critical Metabase patch (CVSS 10.0) landed on 6 August, and Mathspace applied it on 29 August, 23 days later, without running the compromise checks the upstream vendor had recommended for systems patched late. Between the two, attackers accessed the reporting database on 10 August and downloaded 1,079,819 records on 27 August.

Why this matters: for a school business manager or IT director writing the Term 4 vendor risk register, the ASD Essential Eight framing sharpens the number that goes on the council paper. The ed-tech contract that does not name a maximum patch window of 48 hours for critical CVEs, a documented compromise-check obligation when a patch is applied late, and a vendor-side breach notification service level to the school customer, is now the contract that fails a specific published Commonwealth benchmark, not a general "best practice" line. The Cluey Learning, Maths Pathway, Education Perfect, Mathletics, Stile and Corella procurement conversations this term have a hard number to test the vendor answer against, and the 48 hour ASD figure reads directly onto the 72 hour breach clock in the tranche 2 story above. Two calendars, one operational conclusion: the ed-tech vendor whose incident response plan starts on day thirty is running on a clock the regulator has stopped using.

PISA 2025 finds daily AI use hurts scores, weekly AI use does not, and Australia's 15 year olds sit at the high-frequency end

The OECD's PISA 2025 write-up shows students using AI chatbots almost every day for schoolwork scored notably lower in science than non-users, while those using AI once or twice a week performed at roughly the same level as non-users once socioeconomic status is controlled for (Fortune on the OECD's PISA 2025 finding that moderate weekly AI use tracks non-users while daily use trails). 86 per cent of students across OECD countries used AI chatbots for schoolwork in the survey window. The OECD's read is that moderate and intentional use may be positively related to performance and that excessive reliance undermines critical reading and analytical thinking, and Australian 15 year olds reported AI use above the OECD average in the same release.

Why this matters: for an Australian curriculum lead, head of teaching and learning, or director of digital learning writing the 2027 AI-in-classroom policy, the PISA number changes the shape of the guidance. The frontier question is no longer whether students use AI (they do, at above the OECD average), but at what frequency and with what teacher scaffolding. Two implications. First, "once or twice weekly with structured guidance" is now the target use pattern for a teacher-facing rubric, and any 2027 assessment reform that assumes daily student use is on the wrong side of the OECD data. Second, for a NSW deputy principal writing a Term 4 program on top of Prue Car's screen time cap (yesterday's brief) and the NSWEduChat student rollout that opens on 14 October, PISA gives the specific evidence to argue for teacher-shaped weekly use over blanket restriction or unrestricted daily access. A one-page framework the sector can defend in a parent meeting is now available: teacher-designed task, weekly cadence, guided by an instructor who has already used the tool, with the assessment separated from the tool. The tranche 2 privacy bill and the ASD patch clock both make the same argument on the vendor side of that framework.

Four things line up this Thursday. Consultation, before the deal not after. Notification, within seventy-two hours not thirty days. Patching, within forty-eight hours not twenty-three days. And AI use, once or twice a week not every day. Each of these is a number a September or October school council paper can carry, and each has a specific published benchmark to test the vendor answer against. More on the operational landscape at digitalattitudes.com.au.