Every AI restriction announced this term has arrived with the same consolation attached: students will not be using the tools, but they will be learning AI literacy instead.
New York City banned generative AI learning tools for elementary and middle school students on September 2, 2026, a decision Alyson Klein reported in Education Week as reaching roughly 600,000 children. By the end of 2025, according to the NASBE Policy Update written by Winona Hao, 34 states had published statewide guidance on AI in schools, and Ohio had gone further, requiring every district to adopt a formal AI policy by July 1, 2026. The phrase "AI literacy" sits inside a great many of those documents. A sentence telling a head of year what to teach on Tuesday sits inside almost none of them.
That gap is what reaches me. It is one of the questions leaders send in most often through the newsletter, and it almost never arrives as a curriculum question. It arrives as a governance question wearing curriculum clothes: we have committed to AI literacy, and nobody is sure what we committed to.
The Short Answer
AI literacy in a school is not a syllabus. It is three decisions: what your students will still be able to do without AI, where in the week they get to choose whether to use it, and what the school rewards when they choose not to.
Content matters, and the published frameworks supply it well. But every one of them can be satisfied by a school that teaches children to operate AI competently and never once teaches them to put it down. Operation is the part the tools already teach for free, at home, without your help. The scarce thing is restraint, and restraint is a leadership decision before it is a lesson.
Why the Frameworks Cannot Answer This for You
The international frameworks are strong on competencies and silent on the two things a school controls, which are the timetable and the mark scheme.
The most developed of them is Empowering Learners for the Age of AI, the AI literacy framework for primary and secondary education published by the OECD and the European Commission on June 17, 2026. It sorts AI literacy into four domains: Engage with AI, Create with AI, Manage AI, and Shape AI. It treats the literacy as a set of knowledge, skills and attitudes that lets a learner understand how AI systems work, evaluate what they produce critically, and use them ethically. UNESCO's AI competency framework for students, published on August 8, 2024, crosses similar ground through four dimensions, and puts "a human-centred mindset" first, ahead of AI techniques and system design.
Read either document and you will find little to argue with. That is exactly what makes them safe to adopt and easy to leave unimplemented. Neither one can tell you which forty minutes of your week this occupies, who teaches it, or what happens to the student who does it well. Those are not failures of the frameworks. A document written for forty education systems cannot make a timetable decision for one school. It does mean that adopting a framework is the start of the work, and schools keep filing it as the finish.
There is a harder difficulty underneath. What the frameworks describe, at their best, is judgment. Sam Illingworth of Edinburgh Napier University put it about as economically as it can be put, quoted by Sreedhar Potarazu and Carin Isabel Knoop in the American Bazaar on September 3, 2026: AI literacy is "knowing when to use AI and when to leave it alone." I think that is right, and it is the sentence that breaks most school plans, because a student exercising precisely that judgment produces no evidence of having done so. They hand in something rougher and slower than the student beside them who did not.
Decision One: What Students Will Still Do Without AI
The first decision is to name, in writing, what your students will still be able to do without AI once this has worked, and to point at the lessons where each of those things is built.
Almost every AI literacy plan I read describes what students will be able to do with AI. Few say what they will be able to do without it, which is the only half of the pair that can be lost. So write the second list first, and keep it concrete: read something long and difficult without a summary, write a paragraph from a blank page, do the arithmetic, hold an argument out loud with somebody who disagrees. Then find the lessons where each one is practiced and checked, and be honest about whether those lessons still exist in the form you remember.
There is a reason to hold that line more firmly than the marketing suggests. In evidence to the Commons Education Select Committee, reported by Jabed Ahmed in Tes on September 9, 2026, an analysis of PISA data was presented showing students who used AI chatbots scoring around 20 points lower in science than students who did not, while students using them once or twice a week performed about the same as non-users. That second clause is the useful part. What the figures point at is not the tool but the dose, and a dose is a habit, which means it is something a school can teach. It also means the school has to decide what the right dose is, because nobody else in the building is going to.
Decision Two: Where Students Get to Choose
The second decision is to put a real choice in front of students, repeatedly, somewhere an adult can see it, because judgment is only learned where a choice exists.
A school that bans AI has removed the choice. A school that hands out unlimited access has removed it too, in the other direction: a tool available on every task is not a decision, it is a default. Either way students spend the year making no judgments at all, and then leave for a workplace or a university that will offer them nothing but judgments. If you are weighing a restriction, that is the trade-off to weigh, and it is separate from the safeguarding case for restricting, which can be perfectly sound. I have set out how to make a restriction decision you can defend elsewhere.
The alternative is smaller than it sounds. One recurring task per subject where the student decides for themselves whether to use AI, writes a single line saying what they used it for and what they did themselves, and is then asked about that line rather than punished for it. That line is the whole apparatus. It turns an invisible decision into a visible one, which is the only way a teacher can coach it and the only way a leadership team will ever know whether any of this is working.
It is also the thing AI detection cannot give you, and one reason I keep telling leaders not to buy it. A detector tries to catch a decision after the fact and is wrong often enough to do real damage: Potarazu and Knoop point out that detection systems misfire against students writing in a second language and neurodiverse students in particular. Asking a student what they did is simpler, more accurate, and the only version of the two that teaches anything.
Decision Three: What the School Rewards
The third decision settles what students learn, and most schools never make it: whether a student who chose not to use AI, and handed in something rougher as a result, gains or loses by it.
Everything in the first two decisions comes undone here. Students are not naive. If polish earns the mark, they will produce polish, and they will use whatever produces it fastest, and no quantity of AI literacy lessons will outrank that signal. Potarazu and Knoop describe the pattern it produces: stronger grades on written work sitting alongside weaker performance in exams. That school taught AI use through its reward structure and AI literacy through its curriculum, and the reward structure won, as it always does.
Which is why AI literacy is an assessment question before it is a curriculum question. If you want students to learn restraint, some of what you mark has to be work a machine could not have helped with: the live explanation, the redraft done in front of you, the case argued in a room. I have written at more length about what happens to assessment when AI can produce excellent work. The short version is that a mark is supposed to make a claim about a particular student, and a polished product no longer makes one.
The Put-It-Down Test
The Put-It-Down Test is three questions I ask of any AI literacy plan before a school adopts it: What will students be able to do without AI once this has worked, and where would we see it? Where in the week does a student decide whether to use AI, and does anybody ever see that decision? And if a student chose not to use it and handed in something rougher as a result, would this school reward that or mark it down? A plan that survives all three is teaching judgment. A plan that fails any of them is teaching operation, and operation is the one part the tools already teach for free.
It is a filter I suggest rather than a validated instrument, and it takes a leadership team about twenty minutes. The order matters. Teams almost always want to begin with the first question, because naming capabilities feels like curriculum work and curriculum work feels productive. Begin with the third. It is the only one of the three that asks the leadership team to give something up, and if the answer to it is "mark it down", the other two are decoration.
What I See in Practice
Across the leadership teams I work with, the AI literacy conversation nearly always starts in the wrong document: it starts in a scheme of work when it should start in a mark scheme.
The pattern repeats often enough that I now wait for it. A school has adopted a framework, sometimes a good one. Somebody has been handed the job. There are slides. Then I ask what happens to a Year 10 student who writes an essay unaided and turns in something visibly less fluent than the one next to it, and the room discovers it has built an AI literacy program on top of an assessment system that quietly contradicts it. Nobody designed that contradiction. It is what you get when the curriculum decision is made by one group and the reward decision by another, several months apart.
The schools that get furthest are rarely the ones with the most ambitious plan. They are the ones that changed a single visible reward first, often something as small as adding a two-minute verbal defense to one assignment a term, and let the teaching follow the signal. Sitting as a school trustee, I have also watched how fast this question arrives at a governing body, usually dressed as a safeguarding or standards item, and how seldom anyone brings the board the third question. Boards are well placed to ask it, because it is a question about what the school values rather than a question about technology.
The Objection Worth Taking Seriously
The strongest objection is that students do need to understand how these systems work, and that a school reducing AI literacy to restraint will turn out students who are ignorant as well as cautious.
That is correct, and it is why the OECD and UNESCO domains belong inside the plan rather than next to it. A student who does not know that a model predicts plausible text, and cannot say why it will invent a citation with a straight face, cannot evaluate anything it hands them. Understanding is not optional, and it is not the part I would cut.
The argument here is about what the school has to add. The content of AI literacy is now well specified and free; two international bodies have done that work, and a school can adopt it in an afternoon. What no framework can supply is the timetable slot, the visible choice and the mark scheme. Those are the scarce goods, and a leadership team is the only body that can authorize them. If your school has none of the content yet, start with the frameworks. If it has adopted a framework and changed nothing about how it assesses, the problem is not AI literacy. It is leadership.
What to Do Next
Run the Put-It-Down Test over the plan you already have, and start with the third question.
Three practical moves follow from it. Name one assessment this term where a student's own unaided work carries weight, and tell students that is what it is for. Add the one-line declaration to a single recurring task in each subject, and brief staff to ask about it rather than police it. Then write down what students will still be able to do without AI, put a date against a review of it, and give the list to your governors alongside the framework you adopted, because the framework is the easy half and they should see both.
None of that needs a platform, a purchase or a new post. It needs a leadership team willing to decide what it values and then mark accordingly. If you want a wider view of the same terrain, what schools should be preparing children for now sets out the decision underneath this one, and what skills matter when everyone has access to AI covers the capabilities that become scarcer.
If your leadership team is working out what AI literacy should actually mean in your school, that is the kind of work I support through AI strategy sessions with schools and trusts.
Sources and further reading
- Empowering Learners for the Age of AI: An AI Literacy Framework for Primary and Secondary Education, OECD and European Commission, OECD Publishing, June 17, 2026.
- AI competency framework for students, UNESCO, August 8, 2024.
- Sreedhar Potarazu and Carin Isabel Knoop, Who gets to decide what AI literacy means?, the American Bazaar, September 3, 2026.
- Alyson Klein, New York Restricts AI in Elementary and Middle Schools. Will Other Districts Follow?, Education Week, September 2, 2026.
- Winona Hao, States Take Next Steps on Governing AI Use in Schools, NASBE Policy Update, Vol. 33 No. 2, February 2026.
- Jabed Ahmed, Schools "desperate" for clearer AI guidance, Tes, September 9, 2026.
Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and a bestselling author on AI in education. He was a secondary teacher and assistant headteacher before he began advising schools, trusts and governments on how to lead through AI, and he is a school trustee.
