The Take-Home Essay Is Dying: AI and the Integrity Arms Race

by | Sep 1, 2026 | Education in the Age of AI

Skip to main content
< All Topics
Updated
ByThe Good

If there's one place the AI-in-education debate stops being abstract, it's assessment. Every other question — workload, equity, governance — is downstream of a much blunter one: if a student can generate a polished essay in thirty seconds, what exactly is a piece of homework, an assignment or an exam still measuring? Australian schools and universities have spent three years discovering there's no clean answer. And as of this week, the country's largest state has actually landed on a rule — making this the most live, most concrete fight in this entire discourse.

What just happened in NSW

On 10 August 2026, NSW Deputy Premier and Education Minister Prue Car asked the NSW Education Standards Authority (NESA) to urgently examine measures to protect the integrity of the Higher School Certificate, as an immediate step asking NESA to consider a moratorium on unsupervised, take-home assessment tasks entirely. The trigger was concrete: more than 1,270 recorded cheating cases in HSC school-based assessment in 2025, the majority of them tied to take-home tasks. Half of a student's HSC mark comes from school-based assessment, and schools have historically had discretion over how much of that half is done at home versus in a supervised classroom — discretion that's now colliding directly with tools that make unsupervised authorship almost impossible to verify.

NESA reported back on 1 September 2026 — literally the day before this document was compiled — with its actual answer: schools will be limited to a maximum of one take-home assessment task, worth no more than 15% of the school-based assessment component, or 7.5% of a student's total HSC mark. HSC major works — the extended creative and design projects in subjects like Visual Arts and English Extension 2 — are exempt from the cap, though schools must still authenticate that the work is a student's own. The rule applies from the HSC Class of 2027, meaning students starting Year 11 in Term 1 2027 will be the first cohort under the new limit. Car framed the decision bluntly: "when students outsource their learning — whether that is to tutors, classmates, or artificial intelligence — they are missing out on the rigorous, high-quality education they need." NESA's chief executive, Paul Martin, put the stakes even more starkly, warning the consequences of inaction "could be severe over time."

Why this isn't as settled as it sounds

Not everyone thinks capping the format actually solves the problem it's aimed at. AI-in-education researcher Leon Furze, who has reviewed the senior-assessment handbooks of every state and territory curriculum authority, points out that NESA's own published guidance states school-based tasks are designed to measure "a wider range of skills and knowledge" than a formal exam ever could, and that NESA's assessment programs already cap strict exam-condition tasks at one in four, weighted no more than 30% of the total. Stripping unsupervised work out of the other 50% of the HSC, in his reading, cuts directly against the certificate's own design rather than restoring it. There's also a practical question the rule doesn't resolve: if a single remaining take-home task is now worth real marks and everything else moves into supervised, exam-style conditions, does that just concentrate the same integrity pressure onto fewer assessment points, and put even more weight on major works that are explicitly exempted from the cap?

There's a deeper concern sitting underneath the cheating framing, too. University of Technology Sydney's Professor Leslie Loble, an expert in AI and education policy, has argued the implications of AI in learning spaces "go beyond cheating" altogether: learning depends on converting new information in working memory into durable, retained knowledge, and a shortcut around that process may cost students something a plagiarism rule can't capture or restore, whether or not any rule was technically broken.

The same fight, one rung up — universities

Australian universities are living a parallel version of this without a single, coordinated NSW-style rule to fall back on. Turnitin, the plagiarism-detection company whose software sits behind most Australian university submission portals, reported that 53.6% of Australian tertiary submissions run through its system between October 2025 and April 2026 showed some form of AI involvement, with 10% of those more than 80% AI-written. That number needs care — a detection score estimates likelihood, not a proven breach, and can just as easily reflect disclosed, permitted use as misconduct — but a majority-AI-touched submission pool is a genuinely different starting point for assessment design than universities had three years ago.

The technical fix that seemed obvious in 2023 — run everything through a detector — is being quietly abandoned. The University of Queensland turned off Turnitin's AI-detection feature; Curtin followed. The Australian Catholic University reportedly had 6,000 flagged cases, a volume that made false positives an unavoidable part of the process. TEQSA itself described generative AI in a 2024 report as an "evolving risk" — deliberately hedged, because the honest position is the rules haven't settled and vary significantly by institution. The University of Sydney has published a detailed policy spelling out when AI can and can't be used per assessment; the University of South Australia instead walks students through integrity scenarios rather than issuing blanket bans; the University of Western Australia takes a citation-based middle path — AI use is fine where explicitly permitted, but it must be disclosed and referenced like any other source, with the tool named and its role in the work made clear.

What "authentic assessment" actually means in practice

The response taking shape across both sectors isn't better detection — it's assessment redesign. In universities this is described as a shift toward "authentic assessment": automated grading and AI tools absorb the more mechanical, objective parts of marking, freeing staff time for the high-touch feedback and oral or in-person components that are far harder to fake convincingly. In schools, the equivalent is what NSW has just legislated by another name: fewer unsupervised tasks, more oral defences, process diaries and exam-condition work. A national document analysis of Australian university AI policies published in 2026 found something important underneath the redesign trend, though: most universities have introduced AI policies, but there's a real gap between what the policy says and what's consistently applied at the level of a given unit or marking rubric.

The underlying disagreement

Strip away the mechanics and there's a genuine values disagreement sitting underneath the assessment-redesign consensus. One camp treats this as primarily a fraud-detection problem: the tools changed, so the safeguards need to change to match, and the goal is restoring confidence that a grade still certifies what a student can do unaided — which is the frame NSW has explicitly adopted. The other treats it as a chance to ask why so much of what's assessed could be outsourced to a language model in the first place. Both camps now agree the take-home essay in its old, unsupervised form is not long for this world in NSW at least. They don't agree on what should replace the thing it used to measure, or whether a percentage cap is a genuine fix or a politically convenient placeholder.

The question worth sitting with

NSW has just made the most concrete, government-mandated call in the country on this question — and done it in three weeks, from ministerial request to finalised policy. Every alternative on the table — supervised assessment, disclosure-and-citation models, redesigned "authentic" tasks — trades away something the old system had: flexibility, take-home learning time, or the efficiency of setting one assignment for two hundred students. Has NSW actually solved its integrity problem, or just relocated the pressure onto the one task still allowed home and the "major works" that remain exempt from the cap altogether?

Sources

Table of Contents