Before an Australian teacher uses generative AI with students, they need to know four things: what their own system or sector has approved (it differs by state and between public, Catholic and independent schools), what the national Framework actually requires of them, where student data goes when it enters a tool, and how to set a task so that AI use is visible rather than hidden. The technology is the easy part. Most of the trouble teachers have run into since 2023 has come from one of those four, not from the model.
This piece is for classroom teachers and school leaders. It is not legal advice, and where your department's policy is stricter than anything here, the policy wins.
What does the Australian Framework for Generative AI in Schools require?
That AI use in schools be guided by six principles — teaching and learning, human and social wellbeing, transparency, fairness, accountability, and privacy, security and safety — and that schools be able to show how they apply them. The Framework was agreed by all education ministers in late 2023 and is national; it is not a product approval, a curriculum or a ban. It puts the decision about which tools with which students in the hands of each jurisdiction and sector, which is why the practical rules differ by state.
For a teacher, the Framework translates into three questions to be able to answer about any AI use: does it serve a learning purpose you can name, do students and parents know it is happening, and is the tool's handling of student data acceptable to your system. If you can answer all three, you are inside the Framework. If you cannot answer the third, stop.
Which AI tools are teachers allowed to use in Australian schools?
The ones your department or sector has approved for your school, and the list is shorter than the internet suggests. Public systems have taken different paths: New South Wales built and trialled its own tool for public schools rather than approving consumer products; South Australia trialled an education-configured version of a commercial chatbot; other states have approved specific vendors under their own data agreements. Catholic and independent schools set their own policies, often per school. In every case, a consumer chatbot on a teacher's personal account is generally not the approved route for student use, even where teachers may use it for their own planning.
Check your system's current list rather than a friend's, because these change each term. The safe default while you find out: teachers can use AI for their own preparation with non-identifying material; students use only what the school has switched on.
Where does student data go, and what must never be entered?
Into the vendor's systems, under whatever terms your school signed — and under consumer terms, potentially into training data. The Framework's privacy principle and the Privacy Act both apply. The rule that keeps you safe is short: no student names, no work that identifies a student, no assessment records, no health, welfare or family information, and nothing about a colleague, in any tool that has not been approved for that data. "Anonymise it first" is a reasonable habit for planning work; it is not a licence for student data in an unapproved tool.
Teachers are also the front line for a second issue: students entering their own personal information into tools. That belongs in the lesson where AI is introduced, not left to chance.
How do I set tasks so AI use is visible rather than hidden?
Design the task so that the process is what is assessed, not only the product. The assessment problem — a polished essay that a student may or may not have written — is real, and detection tools do not solve it; they produce false positives on the students least able to contest them. What works is changing the task:
- Assess the draft trail. Plan, first draft, feedback, revision. AI can help at every stage and the trail shows how it was used.
- Ask for the conversation. Where AI is permitted, the transcript is submitted with the work and the student annotates what they kept, changed and rejected.
- Move the demonstration into the room. A five-minute conversation about the work, a short in-class task, a presentation. Nobody can outsource what they say to your face.
- Grade judgement about the tool. "Here is what the model said; what is wrong with it and how do you know" is a task that gets harder with AI, not easier.
Assessment in the age of ChatGPT and teaching with AI: a practical guide for educators go deeper on task design.
What should students be taught about AI itself?
Three things that a fourteen-year-old can understand and most adults do not: a model predicts likely text rather than looking things up, so it can be fluently wrong; it only knows its training data and what you give it, so it is a reasoning partner, not a source; and anything typed into it may be stored. Teach those and you have taught the core of AI literacy. The debate about whether AI literacy should be a formal subject continues — AI literacy: should it be a core skill in every school covers both sides — but the three ideas above do not need a curriculum change to teach this term.
How much time should a teacher put into learning this?
Ten to fifteen hours, mostly using the approved tool on your own planning work, and then a term of practice. The productive first uses are the ones that give you time back — differentiated versions of a text, a rubric drafted from your criteria, a unit plan outline, feedback comments drafted for you to edit — and they teach you the tool's failure modes on low-stakes material before you put it near a student. AI fluency for educators is the longer treatment.
If you want a graded, structured version, the AI for Teachers course is built around producing those artefacts for your own classes, and the free AI for Teachers skill check takes three minutes with a free student account and tells you where to start.
The short version to pin above the desk
- Approved tools only, for students. Your own planning: approved tools, non-identifying material.
- No student names, work, records or welfare information in anything unapproved.
- Tell students and parents when and how AI is used.
- Assess the process, or assess in the room.
- Teach the three ideas: predicts, doesn't know, may be stored.