What a Good AI Curriculum for Secondary Students Should Include

Quick answer

A good AI curriculum for secondary students is built around five sequenced capabilities - Understand, Use, Evaluate, Build and Lead - taught age-appropriately from Year 7 to Year 12, with every stage tied to an assessable, portfolio-ready artefact. It is not a chatbot tutorial and it is not a coding elective. It teaches how AI works and where it fails, how to direct it and judge its output, how to use it honestly, and how to apply it to real problems with growing independence. The structure should align with the UNESCO AI Competency Framework for Students, sit inside the Australian Framework for Generative AI in Schools, and assess the thinking - not just the polished output. Done well, it produces a young person who can use AI to go further while still being able to think without it.

Why this matters now

The behaviour has outrun the curriculum, and the gap is now a strategic risk rather than a teething problem. An Elevate Education survey of Australian high-school students found roughly three-quarters now use AI at least a few times a week and almost a quarter use it daily, with ChatGPT the most common tool. The tools are in every classroom and bedroom; what is missing is a deliberate sequence that turns casual use into capability. A school that leaves that sequence unwritten is not staying neutral - it is outsourcing the design of its students' AI habits to whichever app they opened first.

The policy scaffolding, by contrast, is well advanced, which removes the usual excuse for delay. The Australian Framework for Generative AI in Schools - six principles spanning Teaching & Learning, Human & Social Wellbeing, Transparency, Fairness, Accountability, and Privacy, Security & Safety, supported by 25 guiding statements - was approved by Education Ministers in October 2023, and its 2024 review was endorsed again by Ministers in June 2025. ACARA's Australian Curriculum Version 9.0 renamed the old ICT Capability to Digital Literacy, broadening it from the "how" of tools to the "why" and "when", with a sharper focus on privacy, security and online safety. A mandate for digital judgement already exists in the national curriculum. What most schools lack is the AI-specific scope-and-sequence that operationalises it.

The commercial stakes are now explicit in the national accounts, not just in the staffroom. The Productivity Commission's August 2025 interim report, Harnessing data and digital technology, estimated that AI could add roughly $116 billion to GDP over a decade and lift labour productivity by about 4.3%, and urged a growth-focused posture toward AI rather than a precautionary one. The Tech Council of Australia and Microsoft put the generative-AI prize at up to $115 billion a year by 2030, equivalent to two to five per cent of GDP. Those are not classroom numbers; they are the economy the current Year 8 cohort will graduate into. A school designing an AI curriculum is deciding how well its students will participate in that value - as people who direct the technology or as people directed by it.

The labour-market signal sharpens the point. PwC's 2025 Global AI Jobs Barometer found that roles requiring AI skills carry a 56% wage premium, more than double the prior year, and that those jobs grew 7.5% even as overall postings fell 11.3%. The World Economic Forum's Future of Jobs Report 2025 puts analytical thinking as the single most-valued core skill, AI and big data as the fastest-growing one, and estimates 39% of workers' core skills will change by 2030. A good curriculum is not teaching a passing tool; it is teaching the judgement the next decade has already priced.

What a good AI curriculum really means

A good AI curriculum is a structured progression of judgement, not a list of tools or tricks. It rests on one principle that should be visible in every unit: AI should extend a student's thinking, never replace it. Everything in the sequence either builds that judgement or makes it assessable.

That principle has three practical consequences for design. First, it is mostly embedded, not a bolt-on subject - the durable learning happens when students apply AI to real English, science, HASS and arts tasks, with a thin spine of explicit teaching to establish shared language and ethics. This is not a stylistic preference; it is what the learning science recommends. The Education Endowment Foundation, whose evidence is translated for Australian classrooms by Evidence for Learning, finds that teaching metacognition and self-regulated learning is among the highest-impact, lowest-cost strategies available - worth roughly seven months' additional progress a year - and that it works best when taught alongside subject content rather than as a standalone unit. AI judgement is metacognition with a new object; the same evidence applies.

Second, it is sequenced and cumulative: no capability is skipped, and each year deepens the last rather than repeating it. The UNESCO AI Competency Framework for Students (2024) gives this its backbone - a deliberate beginner-to-advanced progression across a human-centred mindset, ethics, AI techniques and AI system design - and it exists precisely because ad-hoc AI instruction tends to teach the same shallow "how to prompt" lesson at every year level and call it progression.

Third, every stage produces an artefact - a finished piece of work plus the thinking behind it - so capability can be seen, assessed and carried into a portfolio. Here the curriculum borrows from two well-established pedagogies. Harvard Project Zero's Teaching for Understanding defines understanding not as recall but as a flexible performance - the ability to apply, explain and extend an idea in a new situation - which is exactly the standard an AI-saturated classroom needs, because recall is the one thing a chatbot can fake. PBLWorks' Gold Standard project-based learning adds the insight that durable capability comes from making authentic artefacts for a real purpose and audience, not from completing exercises. Together they answer the question every head of teaching and learning eventually asks: if AI can produce the output, what exactly are we assessing? The thinking and the performance - and you assess those by demanding visible, authentic work.

The five capabilities: a scope-and-sequence

The spine of the Edison Method is five capabilities, taught in order and never skipped: Understand → Use → Evaluate → Build → Lead. The sequence is the curriculum. Younger students live mostly in the first capabilities; senior students reach the last. The table below is a real, year-by-year scope-and-sequence a secondary school can adopt or adapt.

The point of the sequence is that judgement compounds. A Year 9 student who has already logged where AI was confidently wrong (Year 7) and practised directing it (Year 8) is ready to evaluate output critically. A Year 12 student who skipped those steps is just a faster typist with a plausible-sounding blind spot. Each capability maps cleanly to the UNESCO progression - from a foundational human-centred mindset toward AI system design - and gives a school a defensible answer to the question a parent, an auditor or an inspector will eventually ask: "what, exactly, are we teaching, and when?"

There is a market reason to insist on the upper rungs. Stanford HAI's AI Index 2025 reports that two-thirds of countries now offer or plan K-12 computing education - roughly double the figure since 2019 - so basic exposure is becoming table stakes, not a differentiator. The same report notes that 81% of US computer-science teachers believe AI should be part of foundational education, but fewer than half feel equipped to teach it. That gap between conviction and capability is the real constraint on AI curriculum, and it is why a scope-and-sequence has to be paired with teacher development rather than shipped as a document.

How to assess it: the AI Assessment Scale

Assessment is where most AI curricula quietly fail, because they assess the output and the output is the easiest thing to outsource. The fix is to set the rules per task and assess the thinking. The AI Assessment Scale (AIAS) - developed by Perkins, Furze, Roe and MacVaugh, with a refined v2 in 2024 - is the cleanest instrument for this: a five-level spectrum running from "No AI" to "Full AI" that lets a teacher specify, for each individual task, how much AI use is appropriate, and make that level transparent to students.

The scale does two things at once. It removes the ambiguity that drives both anxiety and dishonesty - students know exactly what is allowed for this task - and it lets a curriculum deliberately sequence AI use, demanding "No AI" while a skill is being built and "High" once it is secure. This is the operational counterpart to the EEF's point about self-regulation: you cannot ask a student to regulate their AI use if the expectation is left unstated, and you cannot build judgement if every task is "Full AI" by default. Crucially, the curriculum should assess the process and judgement: the brief, the first unaided attempt, the prompts, the verification, and a short reflection on what was kept, changed or rejected. That reflection is not a nicety - it is the Project Zero principle of making thinking visible, turned into an assessable object. It is what makes capability legible and integrity defensible, a theme we develop in introducing AI literacy without compromising academic integrity.

Three worked examples

Theory is cheap; the curriculum lives in tasks. Each example below follows the same five-part shape: what the student does today, how AI assists, what the human must verify, the learning outcome, and the control that keeps the student in command.

• Year 8 - directing AI in English (AIAS: Mid). A student plans a persuasive essay, drafts their own thesis, then uses AI to surface counter-arguments. How AI assists: it stress-tests the position. What the student must verify: that each counter-argument is real and fairly represented, not invented. The learning outcome: sharper, better-defended reasoning - analytical thinking, the WEF's top-ranked core skill, practised on a real text. The control: the thesis and the prose stay theirs; the prompt log goes in the portfolio.
• Year 10 - an authentic project in science (AIAS: High). A student designs an investigation into local water quality and uses AI to scaffold an unfamiliar statistical method. How AI assists: it explains the method three ways and drafts an analysis approach. What the student must verify: the method is appropriate and applied correctly to their real data. The learning outcome: a genuine PBLWorks-style artefact built on understanding, not borrowed competence. The control: the data, the design and the conclusions are the student's, documented end to end.
• Year 12 - a capstone in HASS (AIAS: High, with disclosure). A student produces an original research piece on a contested historical question, using AI to map historiography quickly before going to primary sources. How AI assists: it builds a fast scaffold of the debate. What the student must verify: every claim, against a real source, because AI fabricates citations with complete confidence. The learning outcome: a portfolio capstone defended orally - understanding as a flexible performance, exactly as Project Zero defines it. The control: a disclosure statement and an oral defence make the student's own thinking visible and assessable.

In each case AI raises the ceiling without lowering the floor - and each task produces an artefact that shows both the outcome and the judgement behind it.

How to implement it

A scope-and-sequence does not need a wholesale timetable change, and the schools that succeed treat it as a layering exercise rather than a rebuild. The realistic path is to add it to what already exists, in this order.

1. Adopt a shared spine. Establish the five capabilities and one assessment scale (AIAS) as common language across faculties, so a Year 9 task means the same thing in English as in science. Without a shared spine, each faculty improvises its own, and the progression the UNESCO framework calls for never accumulates.
2. Map to existing units. Find the authentic tasks you already set and tag each with an AIAS level and a target capability. Most of the curriculum is already there; it needs framing, not replacing - which is also why this is affordable, in line with the EEF's finding that the highest-impact strategies here are low-cost.
3. Build teacher capability first. Teachers cannot assess judgement they have not practised, and the Stanford HAI data shows this is the binding constraint everywhere: conviction is high, readiness is not. Short, practical professional learning is the highest-leverage investment - the gap is rarely the students, as we explore in the role of teachers in an AI-enabled classroom.
4. Make process visible. Require the brief, the first attempt, the prompt log and a reflection as standard artefacts. This is the single change that does most for both learning and integrity, because it makes thinking the assessed object.
5. Build the portfolio from Year 7. Treat every assessable artefact as a deposit in a growing portfolio, so capability accumulates visibly across six years rather than being re-proven each term. A documented portfolio is also what the post-school market increasingly wants to see, not a transcript of grades a chatbot could have earned.

Common mistakes

• Treating it as a coding subject. Building models is useful for a few and essential for almost none; the durable core is judgement. A curriculum that mistakes Python for AI literacy will lose most students and miss the point for the rest.
• Assessing only the output. If the polished product is all that is marked, the curriculum rewards the students who outsourced it - and quietly penalises the honest ones.
• Skipping the early capabilities. A senior student who never learned to spot a confident error is not advanced; they are exposed. The Year 7 "where AI was wrong" log is not remedial; it is load-bearing.
• Banning AI to "protect" assessment. Prohibition drives use underground and ungoverned, teaches no judgement at all, and contradicts the responsible-use direction of the Australian Framework.
• Buying a tool and calling it a curriculum. A login is not a scope-and-sequence. The sequence is the product; the tool is incidental, and next year's tool will be different.
• Leaving teachers behind. Upskilling students while the staffroom guesses simply moves the capability gap to the front of the room - the exact gap Stanford HAI quantifies.

How to know it is working

A good AI curriculum shows up not in tool fluency but in independent judgement that grows year on year. By Year 12, the well-taught student can use AI to attempt harder work and close the laptop and reason unaided; they disclose their use as a matter of habit, catch the machine's errors, and hold views it did not give them. Their portfolio shows a visible arc from "where AI was wrong" in Year 7 to a defended capstone in Year 12 - a flexible performance of understanding, not a stack of outputs. The immature version is the opposite: faster output, thinner understanding, and a dependence no one sequenced or measured.

The recommendation for curriculum leaders is concrete, and it is deliberately not "choose a platform". Start by adopting the five-capability spine and the AI Assessment Scale as shared language, tag your existing authentic tasks against them, make process the assessed object, and build the portfolio from Year 7. That gives you a defensible, evidence-based AI curriculum for secondary students that aligns with the Australian Framework, satisfies the UNESCO progression, sits inside ACARA's Digital Literacy capability, and graduates young people who command AI rather than defer to it - into an economy the Productivity Commission expects AI to reshape by the billions. For the foundations of why this matters, what AI education really means and AI education for teenagers in Australia make useful companion reading.