Digital Duty of Care closes as AI detectors flag Aussie students

Digital Attitudes, 22 September 2026

Tuesday in the September school holidays, and the sector has four hours to lodge on the Digital Duty of Care. Two Monday stories land in the same window: Australian AI detectors flagging innocent students, and UNSW's Andrew Martin publishing a prompt library that treats the chatbot as a coach instead of an answer machine.

The Digital Duty of Care draft closes at noon today, and ed-tech is in the box

The Commonwealth's Online Safety Amendment (Digital Duty of Care) Bill 2026 exposure draft, released 8 September, closes to public submissions at 12 noon today. Feedback is capped at five pages and lodged by email to digitaldutyofcare@communications.gov.au (Holding Redlich on the Digital Duty of Care exposure draft and what Australian businesses need to know before the 22 September closing time). The draft moves Australia from the current co-regulatory Online Safety Act 2021 to a statutory duty on online service providers to take reasonable steps to deliver a safe online environment, with children named as a protected class and harms including harmful content and online bullying explicitly in scope.

Why this matters: for a school AI buyer, a director of digital learning, or a university head of information governance, the four hour countdown is the last window to put a school pattern on the record before the OAIC and eSafety Commissioner interpret the duty against ed-tech products next year. Two operational reads. First, the vendor scope. The definition of "online service provider" covers every LMS, gradebook, wellbeing platform, parent portal, and AI tutor that reaches a K to 12 student. Compass, Sentral, Schoolbox, SEQTA, Canvas, NSWEduChat, EdChat, ClassmAIte, Corella, Mathspace and Khanmigo all sit inside it. Second, the submission gap. The two constituencies that historically under-index on Commonwealth online safety submissions are the independent school peak bodies and the state Catholic education offices, so an AISWA, AISNSW, ISV, CEWA, CSNSW or metropolitan diocese office that has not already lodged has until midday to write down whether gradebook, wellbeing and AI tutor products should carry the same proactive duty as a mass reach social media platform. The sector answer to that specific question is not in the box yet.

Turnitin AI detectors flag innocent Australian students, and NSW cuts take-home HSC to 15 per cent

A Monday 21 September investigation reports that AI detection software used by Australian schools and universities keeps flagging legitimate student work, leaving students to defend their academic integrity against a probability score the vendor's own guidance says should never be the sole basis for action (Mandurah Mail on Turnitin AI detectors flagging innocent Australian students). Year 12 student Gabe Jones was pulled from class over a Turnitin AI score and described the follow-up as "an interrogation". University of Queensland Professor Jason Lodge on the record: "It's making students suffer". NSW will limit take-home tasks to 15 per cent of school-based assessment for the HSC from 2027, and Curtin University disabled Turnitin's AI detection feature earlier this year.

Why this matters: for a NSW head of teaching and learning, a school academic integrity coordinator, or a university associate dean writing a 2027 assessment paper, the piece consolidates two Australian reference points the Term 4 board paper needs to answer against. First, the vendor read. If the LMS or plagiarism vendor still reports an AI score that lands on the student record, and the vendor's own advice is that the score is not sole evidence, the school procurement paper needs a written statement of what the school does with the score, at what threshold, and with what right of reply, before the next tender lands. Turnitin, Copyleaks, GPTZero and Originality.ai all sit in this frame. Second, the assessment read. The NSW 15 per cent take-home cap for HSC from 2027 is now the numeric benchmark other states will be measured against, and it moves the operational load off the AI detector and onto in-class supervised assessment. That is a timetable question, a room availability question, and an invigilator cost question for the school business manager, not a curriculum question. Any 2027 assessment framework that does not name a supervised assessment percentage is a framework that outsourced the AI question to a probability score.

UNSW's Andrew Martin ships an open prompt library that treats AI as a coach, not an answer machine

Writing in EduResearch Matters on Monday 21 September, UNSW Scientia Professor of Educational Psychology Andrew Martin argues that Australian students use AI reactively to complete work, and that shifting them to psychologically informed use is what actually improves learning (EduResearch Matters on Andrew Martin on shifting students from easy answers to genuine learning with AI). Martin's argument centres on Load Reduction Instruction, a five-part sequence covering difficulty reduction, scaffolding, practice, feedback, and independent problem solving. The specific artefact is the GenAI Motivation and Learning Buddy, an open access prompt library covering Claude, ChatGPT, Copilot and Gemini, built on the Motivation and Engagement Wheel's eleven factors. Martin's core distinction, on the record: "How can AI help me become better able to do this?" rather than "How can AI do this task?".

Why this matters: for a head of teaching and learning, a director of digital learning, or a head of English writing a 2027 program against the NSW screen cap (Monday 14 September brief), the Victorian device free primary rules landing Term 1 2027, and the USyd and UNSW ChatGPT Edu rollouts (14 September and 10 September), Martin's paper is the first Australian academic contribution this quarter that ships with a prompt library the school can print and hand a Year 9 student on Wednesday. Two reads. First, the vendor read. A 2027 ed-tech tender for an AI tutor, writing coach, or study buddy product now has an Australian, university authored reference prompt library that a school procurement panel can compare a vendor's in-product prompts against. Reading Eggs, Studyladder, Mathspace, Cluey, Atomi, Studiosity and Education Perfect each need to answer, in writing, whether their in-product prompts are psychologically informed or merely task completing. Second, the assessment read. The distinction Martin draws between "AI does this task" and "AI helps me become better at this" is the same distinction Sydney's two lane assessment structure, UNSW's Can, Can't, Must framework, and Newcastle's orals model each translate at scale. A 2027 assessment framework that carries the Martin distinction on the front page reads back to the students the university has already told they need to learn to use AI critically. Print the library, put it in the Term 4 professional learning agenda, and the Wednesday morning tutorial has a workable script.

Three moves, one Tuesday. Four hours to lodge on the Digital Duty of Care, an Australian AI detector case study for the assessment board paper, and a UNSW prompt library the school can hand to a Year 9 student on Wednesday. More on the operational shape at digitalattitudes.com.au.