Australia has a national rulebook for generative AI in schools. It also has eight different ways of enforcing it. The Australian Framework for Generative AI in Schools sets out six guiding principles: teaching and learning, human and social wellbeing, transparency, fairness, accountability and privacy, security and safety. On paper, it reads like a unified national standard. In practice, however, the framework is policy rather than federal legislation. That single distinction shapes everything that follows. As a result, states, territories, Catholic systems, independent sectors and individual principals each translate the same six principles into their own binding rules. The outcome is a layered governance structure that rewards close reading.
That pattern holds up under scrutiny. In June 2025, Education Ministers endorsed the 2024 review of the framework. The review found it had correctly identified existing challenges and effectively predicted emerging risks, including the increased use of AI deep fakes. Ministers committed to annual reviews to keep pace with the technology. The national layer is holding. Consequently, the more interesting story sits underneath it.
Privacy Rules Travel Down, Then Diverge
The clearest example of the national-to-local handoff is privacy. The framework prohibits training AI models on student data without explicit consent. It also restricts uploading identifiable student records into consumer-grade tools. Furthermore, Ministers backed this with a $1 million commitment for Education Services Australia to update the privacy and security principles protecting students who use generative AI. From there, each jurisdiction builds its own machinery. Victoria published its generative AI policy in June 2024. Under that policy, schools must obtain opt-in parental consent before using any AI tool that requires personal information beyond a basic school email and password. The policy also bans using AI to draft student reports or replace authentic teacher communication.
New South Wales took a different route to the same principle. Rather than regulating access to commercial tools, the department built its own. NSWEduChat is a department-owned conversational AI hosted in Sydney, running in the department’s own cloud environment to keep student data away from commercial platforms. The rollout followed a deliberate staircase. Trials began in Term 1 2024 with 16 schools, then expanded to 50 schools in Term 2. After teachers reported work efficiencies and high engagement, access opened to all staff in Terms 3 and 4. Full student access for Years 5 to 12 followed in Term 4 2025. Both states satisfy the same national principle, yet both arrived at structurally different answers. Multiply that across Queensland, Western Australia, South Australia, Tasmania, the ACT and the Northern Territory, and the patchwork comes into focus.
The Procurement Gap Nobody Legislated
The framework describes what safe AI looks like. Even so, it does not legally require vendors to prove a product works or protects data before it enters a classroom. This gap is commonly overlooked in discussions of the national approach, and it is where much of the real risk management actually happens. Two mechanisms have grown to fill it. At the national level, taskforce working groups are using the framework to develop unified product expectations for third-party EdTech developers. These set out what systems will expect from AI tools before they reach students.
At the operational level, schools lean on the Safer Technologies 4 Schools (ST4S) initiative. ST4S evaluates digital products against a nationally consistent security and privacy framework, developed in collaboration with every Australian state, territory, Catholic and independent school sector. To date, more than 200 products have been assessed through the programme. Together, these form a de facto regulatory layer. Vendors face no legal mandate. They do, however, face a market where systems and schools increasingly refuse to buy anything that has not cleared the bar.
Screen Time Stays Local, By Design
The framework deliberately says nothing about screen time, device limits or age boundaries. That silence is a design choice, and it pushes the most visible decisions down to the most local level. The division of labour looks like this. Mobile phones are governed by state-level policy, with most Australian public systems enforcing “off and away all day” rules. Device and app decisions are made at the school level, where each campus determines whether to run a 1:1 laptop programme, allow Bring Your Own Device, or restrict specific applications. Individual campuses also choose which software enters the building and which gets blocked.
This autonomy has produced genuine divergence. Some schools are building fully integrated digital ecosystems around tools like NSWEduChat. A subset of institutions, Sydney Grammar among them, has instead banned laptops in class to focus on pen-and-paper instruction. As a result, two students in the same city can experience entirely different technology environments under the same national framework. Educators themselves sit in the middle of this. A survey of 250 Australian teachers found that 58 per cent reported improved student engagement with AI, while 60 per cent expressed concerns about its ethical implications. The majority of the profession holds both views at once, which helps explain why local flexibility persists.
What the Layered Model Gets Right, and Where It Strains
The decentralised structure has real strengths. States can move at their own pace. NSW, for instance, could build a sovereign AI tool without waiting for Canberra. Similarly, Victoria could impose strict consent rules without forcing them on jurisdictions with different priorities. Annual national reviews keep the shared principles current while implementation stays adaptive. The strains are equally real. Consistency depends on every jurisdiction maintaining rigour. In practice, the quality of a student’s AI safeguards now depends partly on their postcode. The procurement gap relies on voluntary mechanisms like ST4S rather than enforceable law. Meanwhile, the screen time divergence between digital-first and analogue schools is widening into a question of educational equity that no current policy layer owns.
The Road Ahead
The framework passed its first major test. The 2024 review judged it fit for purpose, and the annual review cycle gives it a mechanism to adapt as the technology shifts. Nevertheless, the open questions sit in the seams between the layers. Whether national product expectations gain enough weight to function like regulation remains unclear. So too does whether consent models like Victoria’s, or sovereign tools like NSWEduChat, become the template other states adopt. Whether local autonomy over devices produces healthy diversity or entrenched inequality is equally unresolved.
For school leaders and policymakers, the practical takeaway is clear. The national framework tells them what good looks like. The hard work of making it real, however, happens in state policy documents, procurement checklists and staffroom decisions, one jurisdiction and one campus at a time.