The Statement Australia Has Been Owed
For three years, Australia has mainly been discussing AI in education in the register of compliance, with the language of permissions and prohibitions, by people forced into reactive postures by a technology that arrived faster than the governance designed to receive it.
The Castlereagh Statement, published in 2026 by a coalition of ‘the willing’, is the first Australian document that refuses this conversation.
I should say at the outset that I signed it. I helped, in a very small way, with its development. So this is not the writing of an outside observer. It is the writing of someone who has been waiting, for some time, for an Australian document that names the question correctly, and who now wants to explain what that document does, what makes it brave, and what will determine whether it matters.
The Statement opens with a structural admission. Australian education and training systems were designed for an industrial economy in which information and cognitive labour were scarce. Age-based cohort progression. Fixed curriculum delivered in fixed timeframes. Credentials as proxies for capability. Multi-year curriculum review cycles. These structures made sense when knowledge was hard to access and the pace of change was gradual. They make less sense now. Generative AI, the Statement observes, is “rapidly matching or exceeding human performance on many cognitive tasks and is advancing at a pace that outstrips multi-year curriculum review cycles.”
This is the diagnostic move. The Statement says, in plain prose, that the system Australia is trying to retrofit AI into was not designed for AI. It was designed for a different theory of what education was for. That theory, in its core assumption, held that the work of education was the transmission of scarce information from those who had it to those who did not. AI does not break that theory. It exposes that the theory was already failing.
I have written elsewhere that AI functions as a diagnostic rather than a disruption. It does not create the fracture lines in education, work or institutional culture. It illuminates fracture lines that were already there. The Castlereagh Statement, more than any Australian education document I have read, accepts this framing in its bones. It does not ask how to defend the existing system from AI. It asks what an education system would look like if it took seriously the world AI is producing.
That distinction is everything.
The Statement builds on three goals and six foundational principles.
The goals are deceptively plain. First, a shared definition of what we value in human educators and learners, with measurement and incentive structures aligned to those values. Second, coherent learning pathways from early childhood through lifelong learning, recognised across sectors. Third, every Australian capable of confidently, critically and creatively engaging with AI.
The plainness is misleading. What sits underneath each goal is a structural ask that most Australian institutions will find uncomfortable.
The first goal asks the question what is education actually for. Not what does it deliver. Not what does it credential. What is it for, as a human activity, in a world where information transmission is no longer scarce. The Statement says the answer should be compassion, curiosity, creativity, collaboration, courage, the metacognitive skills that enable learners to thrive amid change, the capability to learn how to learn. It says these should be the things we measure, reward, recognise and resource. It admits that delivering on this would require transforming “appointment, workload, recognition, and classification models for staff, and syllabus and assessment foci for learners” across every sector. The answer is structural, and the document says so.
The second goal asks for coherent pathways. Not coordination. Coherence. A learner moving from school to VET to higher education to workplace learning should not be repeating themselves or losing recognition for capability already demonstrated. The Statement asks for a connected national compact across sectors that currently treat each other with bureaucratic suspicion. It is a description of an Australia that does not yet exist.
The third goal is more familiar in shape and quietly radical in substance. It is the goal of universal AI capability. But the Statement specifies that this capability is not technical fluency. It is “confidently, critically, and creatively” engaging with AI while “understanding its limitations and maintaining human agency and voice.” That is the language of what I have called the captain’s chair, the posture of directing rather than being directed by, the test of whether the work you produce with AI is work you could still defend if the model failed tomorrow. It is the language of augmentation rather than abdication, written into national education policy.
The principles are where the Statement gets braver.
Principle 1 redefines the educated Australian as one who exhibits “the enduring dispositions and capabilities that we will always value humans exhibiting, even as AI capabilities continue to develop.” That is not a hedge. It is the assertion that the answer to “what should we keep teaching humans” is the answer that survives every plausible AI future. It commits to compassion and curiosity and courage. It commits to deep domain expertise and practical experience as foundations for intellectual honesty. It commits to genuine integration of Aboriginal and Torres Strait Islander perspectives, with explicit reference to Indigenous Cultural and Intellectual Property and Indigenous Data Sovereignty. It commits to discerning partnership with technology as a “highly desirable, and indispensable, part of study, work, and life in the age of AI.”
Principle 2 is a paragraph that should be quoted whenever Australian institutional leaders speak about AI strategy. It calls for institutional humility, “acknowledging that many of the structures, credentials, and pathways we have inherited may no longer serve learners or the community in the AI age.” It says “institutions exist primarily to serve learning and learners.” It says the structures we have inherited “were designed for a different era. Some may not survive in recognisable form, and this represents new foundations to be shaped rather than failure to be resisted.”
Australian institutions are not in the habit of conceding that their inherited structures may not survive.
Principle 3 reconceptualises learning and assessment. It does the work I have been waiting for an Australian document to do. It calls for assessment that draws on “diverse forms of evidence, accumulated over time and across contexts, to verify learners’ capabilities.” This is the longitudinal, capability-based, witness-and-triangulate model that Jason Lodge and his collaborators have been developing through TEQSA’s assessment reform work. It treats assessment as the work of making thinking visible, which is a different problem from detection and a different problem from surveillance. It treats education as relational and grounded in human experience, while accepting that AI may augment learning when used with intentional pedagogical purpose.
It also, in a single line, embraces “desirable difficulties as essential to learning, while using AI to remove unnecessary barriers and improve accessibility.” Desirable difficulties is Robert and Elizabeth Bjork’s term for the productive friction that produces durable learning, the spacing and interleaving and retrieval practice that feels worse in the moment and produces better outcomes over time. The Bjorks have been showing this for three decades.
The Statement says it should be valued, implemented and protected. It separates the friction that builds capability from the friction that excludes learners. The Bjorks from the bureaucracy. That separation is the entire pedagogical argument of the AI moment, and Castlereagh names it.
The implication runs further than the sentence might suggest. If desirable difficulties are the substrate of durable learning, then AI tools that sell themselves on the promise of removing the friction of learning are not aligned with Australian pedagogical consensus. They are aligned with the opposite of it. Every vendor pitch that begins with “we can make learning easier” now has to reckon with a national education document that says easier is precisely what learning often must not be. The Statement does not say this in those words. It does not need to.
Principle 4 is curriculum reform. It asks for shorter review cycles, removal of redundant content, breakdown of siloes between disciplines and between K-12, VET, higher education and lifelong learning. It asks for coherent pathways with robust credit recognition. It asks for developmentally appropriate AI integration with explicit guidance on when, how and why AI should be used at each stage. It treats curriculum as a living thing rather than an artefact reviewed every five years and patched.
Principle 5 is teachers, and it is the principle without which none of the others survive contact with reality. The Statement names the teacher capacity crisis directly. It names workload, casualisation and inequitable working conditions as structural problems that cannot be solved by exhortation. It calls for “genuine workload transformation, empathetic leadership that aligns AI use to institutional values, and sustained professional development that moves practice beyond substitution to genuine transformation.” It says that reward and workload models must be redesigned to value transformational, collaborative teaching work.
Principle 6 is technology. It calls for alignment of technology procurement and deployment with educational values. Data sovereignty. Interoperability. Indigenous data protocols. Participatory procurement involving educators, researchers, students and staff. The model of education as passive consumer of vendor-defined tools is rejected. “Australia’s education sector will collaborate globally and locally to shape ethical and effective AI tools, being mindful that all technologies have in-built values and assumptions.”
These principles, read together, describe an education system Australia does not yet have. They describe what the signatories think it should become.
I want to be honest about why this matters to me, beyond the fact that I signed it.
I have children. They will move through Australian education at exactly the moment the choices in this document are being made or not made. The institutions they encounter will either have done the hard work of becoming worth encountering, or they will have absorbed the Statement into a strategic plan and continued as before. The world my children walk into as adults will be made of humans formed to flourish alongside this technology, or humans formed against it and never told. That is the stake. Everything else in this piece flows from it.
The intellectual reasons run alongside the personal one. For a while, I have been arguing here that the deepest question of AI in education is not technical but human. Not whether AI can produce a passable essay but what a human education is for once cognitive labour is abundant. The arguments have run in several directions: that augmentation and abdication are different relationships with one’s own becoming, that the cognitive science of desirable difficulties tells us friction is the substrate of development rather than an obstacle to it, that detection theatre is a failed response to a deeper question about what assessment is actually for, and that the apprenticeship layer through which junior staff become competent is being eroded by the same tools that promise productivity while the institutions deploying those tools refuse to notice what they are removing.
The Castlereagh Statement, principle by principle, names every one of these arguments and adopts them. These arguments did not originate with me. They have been made by educators, researchers and cognitive scientists for decades, and by writers and practitioners across the AI-in-education conversation for years. The Statement metabolises that work into a single document Australian governments, institutions and employers are now invited to endorse.
This is what consensus documents are for. They take what the field has slowly come to know and translate it into language that can be cited in a strategic plan, used in a budget submission, referenced in a regulatory framework. The work the Statement does is not the work of having the ideas. It is the work of making the ideas official enough to be acted on.
Certain positions are now structurally closed. Detection theatre can no longer claim alignment with Australian educational consensus. Neither can the framing of AI as substitution for student work, nor the vendor pitch that learning should be made easier, nor the assumption that institutions are sovereign over the learners they exist to serve. These positions can and will, still be defended though.
The Statement does something else that is easy to miss in a document this size.
It includes Aboriginal and Torres Strait Islander protocols not as illustration but as governance. The Aboriginal and Torres Strait Islander AI Futures Communique is named explicitly. Indigenous Cultural and Intellectual Property and Indigenous Data Sovereignty are named as principles that AI tools must align with. The document is bookended by phrases gifted by the Muurrbay Aboriginal Language and Culture Co-Operative: Bindaay-girr yam, Marraal juuda-ndi, darruy guunuwaygu, an invitation to work together to build a world worth living in.
This is not Acknowledgement-of-Country decoration. It is the inclusion of Indigenous knowledge protocols inside the question of how AI is governed in Australian education. It is a recognition that the structures the Statement wants to remake have produced specific exclusions for First Nations learners, and that remaking them well requires First Nations leadership in the remaking. The Statement does not promise this will be done well. It commits, in writing, to attempting it.
The action framework that follows is structured around three horizons. The near horizon, urgent stabilisation: phase out detection theatre, embed AI competencies in curricula, set up a national educator capability framework, address infrastructure equity. The medium horizon, structural transitions: realign incentive structures, reduce curriculum duplication, fund educators to spend at least 20 per cent of their professional development time on pilots, establish national networked learning labs, develop educationally purposeful AI tools through consortium models. The far horizon, new foundations: mission-driven and capability-focused learning, lifespan curriculum, seamless movement between formal, workplace and community learning, infrastructure that verifies demonstrated capability across contexts.
These horizons are not promises but a sequenced map of what coordinated national action would look like. The Statement is explicit that it is a green paper, not a plan. It identifies the goals, principles and horizons. It does not specify the architecture. It says the architecture must be built together.
That, too, is a brave choice. Australian education is replete with policy documents that present themselves as completed answers. The Statement presents itself as a beginning.
Now I want to come to the sentence in the document that I find most important, and that is also the sentence most likely to be quietly forgotten by those who treat the Statement as a publication rather than a commitment.
The Statement contains a section called A core contingency for accelerated transformation. It acknowledges that the three-horizon framework assumes a rate of change that allows staged implementation over multiple years. It then concedes that AI advancement appears to be accelerating, and that its impact on society is unpredictable. It identifies trigger signals that would require shortcutting the staged approach. AI capability advancement. Labour market signals. Learner behaviour shifts. International disruption. And it ends with this sentence.
“We commit to serving learners and their learning even if this means eschewing institutional traditions and inertia under the direction of our governing bodies.”
Read it once. Read it twice. An Australian education document, signed by many, has agreed in writing that the institutions they run may not survive in recognisable form. That when the moment comes to choose between the institution and the learner, they will choose the learner.
This is not how Australian education statements usually speak. Australian education statements speak in modernisation, in consultation, in stakeholder engagement, in the future tense about adaptation. They do not commit, in writing, to overriding the inertia.
The contingency clause is not a manifesto against institutions. It is the work of people inside institutions committing, in advance, to a posture that can survive the conditions they expect to arrive. It is a written hedge against complacency. It says, if the slow path proves too slow, we will not wait for permission to act. It is an attempt to take a future possibility seriously now, while the will to do so still exists.
It is the most honest sentence in the document. It is also the sentence that will be the hardest to honour.
The Statement is addressed to Australian institutions. It asks them to display humility about their own structures, to retire credentials that have become proxies, to break siloes between sectors, and to redesign the appointment, promotion and workload models that reward research over teaching. It asks accrediting bodies to permit demonstrated capability in lieu of qualifications. It asks ministers to fund a national body co-governed by schools, VET, universities, industry, community and students, with funding independent of any single department.
The institutions in question are not built for any of this. They are built to protect their boundaries and defend their classifications. They treat accreditation regimes as load-bearing walls rather than choices. The reward structures and workload models the Statement calls on them to redesign are not accidents. They are the mechanism through which prestige flows, careers proceed, research is funded and managers maintain control over their staff. Asking a research-intensive university to elevate teaching is not a request. It is a challenge that will be politely received and deferred to the next strategic cycle.
This is the gap a statement, however good, cannot close on its own.
I have written previously that the actual mechanism through which institutions choose their AI futures is procurement. The Castlereagh Statement comes from a different register, the register of human formation. But statements about formation get politely absorbed into plans while procurement quietly decides the operating reality. A school system can say AI should serve pedagogy, then buy tools that make pedagogy fit the tool. A university can say it values judgement, then deploy systems that reward output speed. A government can sign a statement about formation, then fund the workflows that remove the slow tasks through which junior staff became competent.
Castlereagh knows this, in the way that documents written by people who have spent careers inside institutions know things. That is why the contingency clause exists. The signatories committed in advance to overriding the very institutional traditions they sit inside, because they understood that signing a statement is necessary and not sufficient.
I helped with a small part of this and sat with the people who facilitated it. The Statement is an act of collective courage that is worth taking seriously. It is also a wager.
What I would ask of any reader who has come this far is the same thing the Statement asks. If it speaks to work you are already doing, add your name at castlereagh.ai. If it points to work you have been avoiding, begin. If you sit inside an institution, ask what your institution would have to change for the Statement’s principles to be honoured rather than acknowledged. Detection theatre dismantled. Reward and workload models redesigned. Accreditation regimes opened to demonstrated capability. Indigenous protocols substantively governing rather than ceremonially endorsed. The contingency mechanism triggered when triggering it becomes necessary, not deferred to the next review cycle.
The signatories know what they have promised. The question is whether the rest of the system will let them keep the promise.
There is still time to find out.



Brilliant essay. I have been sharing your recent writings with the editor of the TAKING STOCK newsletter, who has been disseminating enthusiastically.
https://takingstocknewsletter.substack.com/p/taking-stock-humans-in-the-ai-loop?utm_source=post-email-title&publication_id=2249221&post_id=196861932&utm_campaign=email-post-title&isFreemail=true&r=ewwlb&triedRedirect=true&utm_medium=email
A genuinely consequential development and a paradigm-shift framework. Downloaded and queued for later close reading (& sharing). May governments and governance listen and act.