Ask a staffroom what students would put in a school AI policy if you handed them the pen, and most people predict a loosening. Fewer rules on drafting, more tolerance at homework club, and a quiet retirement for the detector software. Over one weekend in Boston, 98 American teenagers were handed the pen, and what they wrote is tighter than most schools would dare publish.
The set-up was a civics exercise. Students from all 50 states, most of them in the last two years of high school, spent a late-July weekend in a full-scale replica of the US Senate chamber at the Edward M. Kennedy Institute, split into subcommittees, arguing over provisions and voting on amendments. The event was hosted with AASA, the American superintendents' association, and Day of AI, a nonprofit that came out of MIT's RAISE initiative. Their Students First Act passed 82 to 16. It has no legal force, though AASA intends to send the text to its 10,000 members, many of whom are writing their own AI rules right now with nothing national to anchor them to.
Here is what the students voted for. No AI for writing, with editing, brainstorming and studying permitted from roughly Year 9 upwards. No AI on any graded test. Any student who is allowed to use it has to say how they used it and be able to demonstrate the learning some other way, in discussion, by hand, or out loud. One student from Mississippi summed up the fear behind the vote: without guardrails, "they're gonna use it as a brain."
They were also explicit about why they moved fast. This cohort watched adults arrive late to smartphones and social media, and they have no appetite for a repeat.
What they asked to be taught
The literacy requirement is the part that should interest anyone doing this work, and it is where the students were most demanding.
They refused to pick a year group. Their solution was to tie AI literacy to hardware: it starts the moment devices appear in the classroom, whenever that is in your setting. The content they wanted covered is wider than most schemes of work I have seen. Misinformation. Plagiarism. Bias in how these systems are built, not only in what they produce. What happens to student data. What counts as legitimate academic use. The environmental cost of running the things.
Read that list as a timetabler and the problem arrives immediately. Six strands, no subject, no time allocation, no assessment. Written like that, it becomes a PSHE carousel or a September assembly, and by February nobody in the building can tell you where bias in AI design was taught, to whom, or whether anyone learned it.
The students got closer to the answer than their own bill did, and they did it by accident. Across that weekend they were evaluating sources, arguing about what counts as your own work, weighing a student's privacy against a school's safeguarding duty, and testing a policy against what schools can actually afford. Every one of those sits on their literacy list. None of it happened in an AI lesson. It happened inside a civics exercise, which is where this material tends to live: inside subjects that already teach the underlying thinking, mostly unlabelled.
That is the argument I keep making to heads of department. You are not adding AI literacy to an already full curriculum. In most cases you are naming what a lesson already does, making the AI dimension explicit, and assessing it. History departments teach provenance. Science teaches sample bias. English teaches voice and attribution. Computing teaches how the model was built. Bolting on a separate unit is what schools reach for when they haven't yet mapped what they've got.
The parts I would take
Detectors get demoted. The bill says a suspicion of misuse cannot rest on a similarity score. Two school officials review the work, and the teacher investigates personally before anything goes to a senior leader. If you read one line of this bill into your own policy, make it that one. Plenty of UK policies still treat a percentage from a checker as evidence, and every head of department who has had that conversation with a parent knows how it ends.
Safeguarding is written into the tool, not the training. The students spent a long time on what should happen when a pupil turns to a school chatbot for emotional support. Several argued for confidentiality, including the real risk that a student could be outed by an automatic alert. They still voted that the tool should stop, decline the request, and bring in an adult. Whatever you think of the conclusion, the procurement conversation in most schools never gets that specific. Ask your provider what their chatbot does at that exact moment, and see how quickly you get an answer.
Somebody asked who pays. When the literacy provision landed, a senator from Tennessee raised what it would cost schools with the least money, and another pointed out that teachers are already stretched. The students patched it with a federal funding requirement, which is optimistic as legislation and honest as policy. It is more than most school AI strategies manage, given how many ask for AI literacy across every subject with no time, training or budget attached.
Teachers publish too. Each term, a written statement of where AI is permitted in that subject and where it isn't. Departmental variation already exists in every school I have worked in. Students are just asking for it in writing, which is cheap to do and moves the burden of clarity off the fifteen-year-old.
Where it wobbles
The line between writing and editing does not survive contact with a Year 10 essay. Brainstorm becomes outline, outline becomes structure, structure becomes topic sentences, and the pupil has written nothing. The provision sounds decisive while leaving the hardest judgement exactly where it already sits, with the class teacher at 9pm.
The ban on AI in graded tests costs nothing. It describes what already happens in every exam hall in the country.
And the cohort matters. These 98 were hand-picked by their school leaders. They are, broadly, students for whom a prohibition on AI writing is a small price, because their writing is already strong. A pupil two years behind in written expression is running a different calculation, and a policy written by confident writers will tend to protect the value of confident writing. That is not a reason to dismiss the bill. It is a reason to run the exercise with a wider group before you borrow the conclusions.
Run the exercise, not the bill
The bill is interesting. The method is the thing worth stealing, and it fits in two lessons.
Give students your current AI policy and ask them to find the sentence that would be hardest to enforce, then argue for a replacement. Ask each department to draft a paragraph on where AI helps and where it gets in the way of the thinking that subject is trying to build. Put the two documents side by side, then let students amend and vote.
Look at what that lesson covers before you decide it is an add-on. Students have to weigh evidence about detector reliability, define academic honesty in their own words, argue about who sees their data, and cost their own proposals. Four of the six strands in the students' own literacy list, taught through a policy review you were going to have to do anyway. Set it as a persuasive writing task and you can assess it in English. Run it in Citizenship or a tutor programme and you have your student voice evidence for the review.
Those teenagers spent a weekend negotiating, amending, defending a position out loud, and changing a vote when the argument moved. The AI literacy was inside the civics lesson the whole time. It usually is.
Source: Cory Turner, "Adults have struggled to set rules for AI in school. These teens figured it out," NPR, 30 July 2026.
AILitKit surfaces the AI literacy already inside the lessons your teachers are planning. Three guides are free on the starter tier.