NYC's school AI ban reaches Grade 8. Singapore starts at P4

Eight days before the school year started, New York City told about 600,000 children that the chatbots were off. A child in a Singapore Primary 4 class this year, meanwhile, can be using an AI tool in a lesson with a teacher watching. Both school systems say they are protecting the same thing: a child’s ability to think.

So which one has it right? The more interesting answer is that they agree on more than the headlines suggest, and that neither rule reaches the place where most of the risk actually sits.

What New York actually switched off

On 2 September, Mayor Zohran Mamdani and Schools Chancellor Samuels announced a one-year moratorium on student-facing generative AI from preschool through eighth grade, about two-thirds of the city’s enrolment. Companion chatbots are banned in every grade. The city’s own announcement lists the exceptions: assistive technology for students with disabilities, tools for multilingual learners, and career programmes such as computer science. Teachers can still use AI, but only for jobs like planning lessons, translating and drafting messages.

The mayor did not hedge. “The tech industry wants us to believe that A.I.-powered early education is not only inevitable, but necessary,” he said. “We do not see it that way.”

This did not come from nowhere. In April, a Panel for Educational Policy meeting ran for nearly seven hours as more than 100 parents, students and educators testified, with many calling for a moratorium.

The fine print is messier than the announcement. The Education Department says it is “disabling” AI features in 38 existing citywide software contracts but has not published the list. The reading tutor Amira and the digital Houghton Mifflin Harcourt reading curriculum will turn their AI components off. Asked whether Google’s Gemini, which comes pre-loaded on many students’ Chromebooks, would be disabled, an official did not directly answer. Chalkbeat notes it is not yet clear how the city will decide whether AI has been “sufficiently scrubbed” from a product, and the guidance does not say how schools should handle AI use outside school.

China and Singapore got there first, more quietly

New York calls its policy the broadest in America. In Asia, similar lines were already drawn, with far less noise.

A Ministry of Education committee on basic education in China issued guidance in May 2025 stating that “primary school students are not allowed to independently use open-ended AI content generators.” Middle schoolers “may explore the logical structure of AI-generated content”, and older students go further into how AI works. The stated worry was children becoming “overly dependent on the technology”. (We looked at China’s wider approach in an earlier piece.)

Singapore’s version came in Parliament on 6 May this year. Education Minister Desmond Lee said that from Primary 1 to 3, schools “will not assign any work that requires them to use AI directly”, keeping those years hands-on. From Primary 4, pupils use educational AI tools under supervision, with low exposure, because by then they have the “executive functioning skills” to plan and check their own work. Part of what they are taught is “when they should not use AI”. In secondary school, students must say where they used AI and cite their sources.

The line they share is not about age

Put the three side by side and the ages differ a lot. New York holds the line until the end of eighth grade. China holds it through primary school. Singapore opens a supervised door at P4.

But the thing being kept away from young children is the same in all three: a child alone with an open-ended machine that will produce the work for them. China’s key word is “independently”. Singapore’s is “under supervision”. New York bans the general tool and then exempts the ones doing a narrow job, like helping a child with a disability read or a new arrival follow a lesson.

There is decent evidence behind that instinct, though it comes from older students. A field experiment in Türkiye, published in PNAS in 2025, gave nearly 1,000 high school maths students either a standard GPT-4 chat interface, a version set up as a tutor with teacher-designed hints, or no AI. With the plain chatbot, practice scores rose 48%, but exam scores fell 17% against the no-AI group once the tool was taken away. The hint-giving tutor lifted practice scores by 127% and showed no measurable harm on the exam. Same underlying model. Different job. (More on that study here.)

What no study tells us is the right birthday. Whether that line belongs at nine, twelve or fourteen is a judgement call, and New York is honest enough to say it will spend the year studying the effects.

None of these rules reach the kitchen table

Here is the gap. Every one of these policies governs school. New York’s guidance leaves outside use unaddressed, and a school rule in Singapore or China cannot follow a child home. In our view, the open-ended chatbot a ten-year-old uses completely on their own is far more likely to be on a phone at 9pm than on a classroom laptop.

That makes the shared line more useful to a parent than any of the ages. A few ways to use it:

• Ask two questions about any AI tool your child uses: are they alone with it, and is it open-ended? Supervised, or doing one narrow job, is the lower-risk end.
• For younger children, keep AI a together activity. They try first, then you look at what the machine says, side by side.
• If an older child uses a chatbot on their own, push it towards hints. Ask it to coach rather than answer, which is the setup that did no harm in the Türkiye study.
• Ask the school which apps on your child’s device have AI switched on. New York has not published that list for its own contracts, so it is a fair question anywhere.

A disclosure: Mentus AI makes an AI mentoring app for children, so we have a stake in how this debate lands. That is partly why we think the ban-or-embrace headline misses the point. The question three very different countries have quietly converged on is not whether a child meets AI. It is whether they meet it alone.