On Sunday evening, the Sydney Morning Herald ran an exclusive claiming NSW is preparing to bar Year 12 students from completing unsupervised take-home assessment tasks, on the advice of a letter the Deputy Premier sent to NESA at the end of July. On Monday morning, the Premier of South Australia announced a royal commission into artificial intelligence, the first of its kind in the country.
Less than a month before that, the Prime Minister stood up at the University of Sydney and announced a national framework covering copyright, data centres, energy, water and the workforce, which I wrote about at the time in Australia Just Set the Terms for AI.
Federal and state ministers, and now the assessment architecture of the country’s largest senior credential. Something is clearly happening. The question is what, and whether the three things are part of the same overarching discourse.
What NSW is doing
The SMH story is built around a letter from Prue Car, Deputy Premier and Minister for Education and Early Learning, to NESA chair Professor Peter Shergold, dated 31 July 2026. It’s worth reading the letter directly rather than the headline, because the two are not saying exactly the same thing.
The headline says take-home assessments will be banned. The letter asks the NESA Board for “urgent advice”, and Car’s main request is this:
“I would be grateful if the Board prioritised advice on a moratorium on unsupervised assessment tasks, with consideration given to differentiating by key learning area where necessary.”
Hardly a “ban on take home assessments”, but given the letter was freely released to the SMH and the article must have been approved, there’s some truth to the headline. Car is explicit that “some measures may need to be interim arrangements while further evidence is developed”, and that advice needs to reach schools before Term 4, 2026, when the next cohort of Year 11 students starts Stage 6.
The letter also contains three specific instructions to the Board that didn’t make the article itself, and I think they’re interesting than the “ban”:
- advice should consider “the evidence base for the creation of a knowledge-rich curriculum as the most effective platform for learning”
- it should consider “how major projects and creative works can be preserved while protecting student authorship and cognitive effort”
- and it should consider “the limited benefit of teaching skills aimed at identifying AI-generated material, given the speed at which AI is evolving”
That last one didn’t make the article at all, but many of us have argued for years that AI detection in education is a dead end. An education minister has put in writing that teaching students and staff to spot AI-generated text is not worth the effort, because the technology moves faster than the detection. It’s good to see it in the letter, because the detection discourse has been a distraction for three years, and the sooner systems stop funding it the sooner they can spend the money on practices that actually work.
The SMH leads with a predictable cheating narrative: 1270 recorded cases of cheating in HSC school-based assessment in 2025, the majority relating to take-home tasks. That figure comes on top of the Learning First report I explored in Exams Are Not the Answer, Part One: a report whose actual recommendation was to review and strengthen the parameters of internal assessment, and which explicitly warned that “a hasty shift to fully in-class, device-free assessment without deliberate design risks stripping away important elements of tasks”. When that report was released, the SMH published a similar shock-and-awe article demanding assessment reform.
Note the pattern. Mainstream media in Australia is already hostile to teachers, and there is also a popular public conversation right now pushing back against “screens” and “devices”. Calls to ban AI that deliberately cherry-pick data and government communications manage to do both things at once: they deprofessionalise educators by conforming to the idea that teachers are lost at sea with new technologies, and they treat technology as the root cause of all our modern ailments.
Conveniently for the technology companies themselves, it shifts all of the focus away from regulating addictive, distracting, data-hoarding and deliberately designed technologies, and instead pushes down the paths of phone bans and age restrictions.
What South Australia is doing
On Monday, South Australian Premier Peter Malinauskas announced a royal commission into AI. According to the press release, the commission will involve three commissioners, recruited through an international search, starting in October with a report due by 1 July 2027. It will cost roughly $3 million AUD and draw on evidence from business, industry, unions, technology developers, academics and the creative industries, covering work and labour, energy and water for data centres, health, the arts, and education.
Malinauskas’ framing will sound familiar if you watched the recent Albanese speech:
“We can choose to allow this industry to come to our country and evolve in a way that is unregulated and unchecked. Or we can choose to set the agenda … to ensure that we use AI in a way that puts people first.”
Peter Malinauskas, 10 August 2026
And, in probably the most quotable line:
“I think AI gone bad will make social media look like a walk in the park.”
This is the same government that ran the state-level campaign on the run up to Australia’s under-16 social media ban, complete with Jonathan Haidt at a roundtable in Adelaide. Malinauskas has run this play before and it worked.
On education specifically, he said he hated the idea of kids graduating from high school having learned how to use ChatGPT but not how to read, write, do arithmetic, or talk to each other. It’s the same anxiety Car is expressing, in almost the same words, two states away.
Malinauskas announced the royal commission days after returning from the United States, where he met OpenAI, Anthropic and Apple, and signed a memorandum of understanding with OpenAI president Greg Brockman; believed to be the first such agreement between an Australian state government and a major AI company. While in the US, Malinauskas also had an “important discussion about artificial intelligence and what the future of the technology could mean for our economy and society” with Apple CEO Tim Cook. A week after meeting senior tech execs in the Valley, the Premier announced the royal commission.
I flagged this tension in July as the unresolved contradiction sitting under the national framework: courting frontier AI investment from the very companies whose practices the standards are designed to constrain.
Two very different instruments
So: is the tide turning against AI in Australia?
Well… no. I think there are many moving parts, but it’s not as simple as the techlash would suggest. The Prime Minister’s concerns that “others will write the rules” and Malinauskas’ royal commission appear to be pulling in the same direction, but the level of involvement of US technology companies should concern us.
NSW and SA have also reached for opposite methods: in South Australia, it’s find out first and then act. In NSW, it’s act now, or more specifically, “act firmly in the interests of students and not allow practices to become established before their impact on learning and adolescent development is properly understood.”
Obviously you can’t run a royal commission on the timeline of a Year 12 cohort: the students starting Stage 6 in Term 4 don’t have the luxury of waiting until July 2027. But “interim” measures in senior secondary assessment have a way of becoming the architecture. Schools rebuild assessment programs, write new task notifications, run PD, and tell parents and boards. Nobody rolls that back that after eighteen months because a commission in a different state recommends nuance.
Unfortunately, the measure itself defies the design of the system it’s supposed to protect. I went through the handbooks of every senior secondary authority in the country for Exams are Not the Answer: Part Two, including NESA’s, which tells students directly that “school-based assessment tasks can measure a wider range of skills and knowledge than an external exam can”, and whose published assessment programs cap formal exam-condition tasks at one in four, weighted no more than 30%. Removing unsupervised tasks from the other 50% of the HSC doesn’t support that design.
What I’ll be watching
Anyway, may you live in interesting times…
I’m 45 minutes over the border from South Australia, and I work with schools and universities in every state and territory. I’ll be watching the royal commission with great interest, but particularly keeping an eye on whether the findings challenge US technology companies or simply fall in line with their current narrative of “AI is unsafe and only we can save you.” I’ll also be interested in these specific questions:
Whether SA’s terms of reference include schools, and how. Whoever drafts the terms of reference in the next four to six weeks can decide whether a royal commission asks “how do we stop students cheating” or “what is school for now that AI exists”. Those would produce completely different reports.
Whether SACE gets dragged along. South Australia runs the most coursework-heavy senior system in the country: 70% school-assessed in every Stage 2 subject, with English and Modern History requiring no directly supervised coursework anywhere in that 70%. In 2025, in the middle of the panic, the SACE Board removed its last closed-book coursework requirement. If SA follows NSW, it has a great deal further to fall than NSW does.
Whether NESA’s advice differentiates. Car asked for differentiation explicitly “by key learning area where necessary” and the SMH reports exceptions are expected for subjects with major projects. Whether that means a handful of carve-outs for Visual Arts and Design and Technology, or a genuinely subject-by-subject reading of where unsupervised work carries the learning, is the whole ballgame. As an English teacher, I’d love to know what extended writing (already not valued in most senior curricula) looks like when everything has to be done in 45-minute supervised blocks of time.
Whether the federal framework and the state instruments end up talking to each other. The national framework goes to National Cabinet this month. A state royal commission reporting in July 2027 into a domain that a federal bill will have legislated on by early 2027 is either useful redundancy or a jurisdictional mess. I guess only time will tell which.
The window of critique
Before these announcements, I wrote on Sunday morning that the window of critique for AI is closing fast: the messy, exposed, obviously-dangerous-looking phase of a technology is the phase in which it can still be interrogated, and once it works well enough nobody remembers what the fuss was about.
A royal commission with the authority to compel evidence from technology companies, and a minister willing to take big swings at assessment practices are the kinds of things that only happen while the wiring is still exposed. Both of those things happened in Australia in the space of twelve hours.
But again, how much of these conversations are being driven by educators, students, and the Australians using the technologies, and how much influence is the technology industry exerting? It might seem counter-intuitive, but big US tech benefits enormously if we all play along with the myth that AI is inevitable and inevitably risky.
If OpenAI, Apple, Microsoft, Google, Anthropic and co. can persuade our government that only they have the power to protect us, then Anthony Albanese’s push to set the terms of engagement for Australian AI has ended before it starts.
Want to learn more about GenAI professional development and advisory services, or just have questions or comments? Get in touch:
Leave a Reply