New York City's new school policy sounds like a ban until you read the rest of it. Student-facing generative AI is under a one-year moratorium from pre-kindergarten through eighth grade. Companion chatbots are prohibited at every age. At the same time, high-school students will have two AI literacy sessions during the year and some will take part in closely supervised pilots.
The policy affects nearly 600,000 pupils. The older students' lessons are two 45-minute modules covering what AI is and is not, bias, risks, ethics and future skills. The pilots are limited to five classes per high school, with vetted tools and teacher oversight.
There is a distinction here that schools often miss. A pupil can learn about AI without being asked to use a chatbot to complete their work. Equally, giving a pupil access to an AI tool does not mean anyone has taught them to judge it.
The age question is really a learning question
New York's boundary is blunt: younger pupils will not have student-facing generative AI in school under the general rule for 2026–27. Its stated aim is to protect foundational learning and give the city time to examine the effects of the technology.
The policy also includes exceptions for assistive technology, multilingual learners and some career-readiness programmes. That matters. A blanket rule that removes a useful support from a pupil who needs it would have a different cost from limiting a general-purpose chatbot used for homework.
For a school considering its own rules, the first question is what the pupil is meant to learn. If the task is to practise composing a paragraph, an instant generated paragraph may defeat it. If the task is to evaluate a claim, a generated paragraph might be exactly the material to critique. The same tool can either support or bypass the learning, depending on the task.
The pilots are small on purpose
The city has named specific uses: close reading, maths coaching, guided practice, examining AI outputs and projects about community problems. Time limits vary by tool. Teachers must supervise, and the city says vendors were selected against safety and privacy criteria.
That does not prove the pilots will improve learning. The coalition reviewing the policy still has a year's work ahead. But it creates a testable proposition: in this subject, for this task, with this amount of use, did pupils learn more or think more clearly? That is far more useful than asking whether a school is “for” or “against” AI.
The literacy provision is thinner. Two sessions can introduce vocabulary and risks; they cannot by themselves make every high-school student a capable evaluator of AI output. The real work will happen when subject teachers ask pupils to check an explanation, justify a source or show which part of an answer was theirs.
What UK schools can take
Write two separate columns in your policy. In one, set out which pupils may use which tools, in what circumstances and with what safeguards. In the other, set out what every pupil must understand about AI by the time they leave. Do not let the second column disappear because the first one is strict.
New York has put a boundary around access and created a space for instruction. The year ahead will show whether the instruction grows into something more substantial than two lessons. The principle is already useful: you can restrict a tool and still teach the literacy required to live with it.
Source: New York City Mayor's Office, school AI policy announcement, 2 September 2026.