New York's schools are right to slow-walk AI for the youngest students

New York City has banned a great many things from its classrooms over the years — cellphones, soda, snow days without remote instruction. This week it banned something the rest of American education has been racing to install: generative artificial intelligence, for six hundred thousand children in pre-kindergarten through eighth grade. It is the right call, and other big districts still rushing AI vendors into their classrooms should be paying closer attention than they are.

A reversal, and a reversal of a reversal

On September 2, Mayor Zohran Mamdani and Schools Chancellor Kamar Samuels announced a one-year moratorium on student-facing generative AI for students from 2-K through eighth grade, alongside new screen-time limits: no individual devices for children through second grade, a 30-minute daily cap in grades three through five, 45 minutes in middle school. The city says it is disabling AI features across 38 existing software contracts, including a reading tutor used in 222 schools and Google's Gemini assistant on student Chromebooks. High schoolers get a narrower carve-out — five supervised pilot programs reaching up to 50,000 students, plus twice-yearly lessons on how the tools work and where they fail. The full particulars are in the education department's new guidance, and by scope alone — the nation's largest school system, two-thirds of its enrollment — this is the most sweeping restriction any American district has placed on classroom AI.

It is also, pointedly, a reversal of a reversal. New York banned ChatGPT within two months of its public release in January 2023, citing worries about accuracy and cheating, then rescinded the ban that May, with the chancellor at the time calling the original ban "knee-jerk fear" and pledging to bring AI into the curriculum instead. An AI Policy Lab followed. Three years later, the city has landed back roughly where it started, only this time the caution is not reflexive. It comes with a paper trail.

"Children need teachers and human connection in order to learn and grow." — Mayor Zohran Mamdani

The evidence the caution rests on

Skeptics of the moratorium will note, correctly, that no rigorous study has proven generative AI tutoring harms eight-year-olds. But the argument for delay was never that AI in schools is proven dangerous. It is that AI aimed at children is largely unproven at all, and the evidence that does exist about how kids actually use these tools is not reassuring. A national census by the nonprofit research group Common Sense Media, surveying children ages nine to seventeen, found that 86 percent now use AI in some form and nearly a quarter do so daily, while more than four in ten say no parent has ever talked with them about its risks. A separate survey from the same group found that 72 percent of teens have used an AI "companion" product, with roughly a third saying they find those conversations as satisfying as talking to real friends.

That matters because regulators are already worried about exactly this category of product. In September 2025 the Federal Trade Commission opened a formal inquiry into AI companion chatbots, ordering seven companies — including Character Technologies, Alphabet, Meta, OpenAI and xAI — to disclose how they test for and limit harm to minors. The inquiry followed wrongful-death lawsuits against Character.AI brought by families of teenagers who died by suicide after developing intense relationships with its chatbots; the company and Google agreed in January to settle several of those cases. None of that is proof that a district-approved reading tutor behaves like a companion chatbot. But it is proof that the industry building these products has not yet earned the presumption of safety that "innovation" arguments usually ask schools to extend it.

A policy with real gaps

None of this makes the New York policy beyond criticism, and the loudest critics are not the ed-tech industry. The teachers' union, the United Federation of Teachers, has warned that the guidance leaves compliance largely to individual schools to sort out, with no clear enforcement mechanism spelled out for a system of 1,600 buildings. A coalition of parents that had pushed for a full two-year moratorium called the policy "a step in the right direction" but flagged the exemptions — for students with disabilities, English-language learners, and the high-school pilots — as the likely pressure points where the ban will erode fastest. Reporting on the rollout by Chalkbeat New York found the city has not published data on how extensively AI tools were already embedded in classrooms before the ban, which makes it hard for anyone outside headquarters to verify the moratorium is being followed rather than quietly ignored. Those are legitimate, specific complaints, and the city's promised oversight body, a Technology in Schools Coalition tasked with reporting back over the course of the school year, needs to answer them with data, not press releases.

The case for making schools the last adopters, not the first

Even with those flaws, New York's basic instinct is the correct one, and it cuts against the prevailing mood in American education policy. As an analysis of the policy noted, most large districts spent the past two years doing the opposite of what New York just did: signing agreements with AI vendors, training teachers to use chatbot-assisted instruction, and adding AI literacy requirements while leaving student access largely open. Los Angeles, Chicago, Houston and thousands of smaller systems have treated AI adoption as inevitable rather than optional, on the theory that children who don't learn the tools now will fall behind children who do.

That theory has it backwards. Schools are one of the few institutions with genuine leverage over how a technology reaches children — leverage that evaporates once a tool is already embedded in every classroom and vendor contract renewal becomes the status quo to defend. Used well, that leverage means schools can wait for companies to answer the FTC's questions, for the pending lawsuits to run their course, and for independent researchers rather than the vendors themselves to say what these tools do to a developing reader or a lonely twelve-year-old. Being the last major institution to adopt an unproven consumer technology aimed at children is not backwardness. For a school system responsible for six hundred thousand kids, it is exactly the caution the job requires — and the districts still racing to get ahead of the evidence, not the one that paused, are the ones that will have to explain themselves later.