3 SEP 2026 — New York City has paused student-facing generative AI for the 2026-27 school year. The moratorium covers children from 2-K through eighth grade, roughly 600,000 pupils, about two thirds of the largest school district in the United States. The part that reaches every grade is missing from most coverage: companion chatbots are prohibited across the whole district, high schools included, and that is the strictest line in the policy.
What the policy actually says
The moratorium runs for the 2026-27 school year. For 2-K through eighth grade it applies to all software using student-facing generative AI, with named exceptions for assistive technology for students with disabilities, tools for multilingual learners, and career-readiness programmes.
High schools get a narrow permit rather than an exemption. Five pilot programmes are allowed: Quill for English language arts, Edia for mathematics, Brisk Teaching for teacher-created activities, and Playlab and Intel AI-Ready Schools across multiple subjects. The cap is 50,000 students in total and five AI classes per school. Every high school student gets two 45-minute AI literacy modules a year.
Teachers may continue using AI for lesson planning and administrative work where the tools meet the city's safety standards. A coalition of educators, parents, elected officials, union partners and outside experts will review the evidence and recommend what happens next.
Moratorium and ban are different instruments
A moratorium with a fixed term, a named review body and a remit to recommend is a suspension pending evidence. The word ban carries a finality this policy does not have, and the difference shapes what happens in September 2027.
Chancellor Samuels framed it that way directly: "Innovation does not mean more technology, and over the next year, we will lead with evidence." Mayor Mamdani's framing was about what the year is for rather than what it forbids. "This moratorium is a commitment to getting the future right. We will embrace new technology, but only when it serves our students."
Reading it as a ban also obscures the mechanism the district is relying on. Nothing here depends on detecting whether a fourteen-year-old used a chatbot at home. The district's lever is procurement: which tools it buys, approves and permits on its own network. That is the only lever a school system controls.
The companion chatbot line is the substantive one
Grade-limited restrictions on homework tools are common enough that a dozen districts have some version. A district-wide prohibition on companion chatbots, applying to seventeen-year-olds as much as to six-year-olds, is a different claim.
Companion chatbots are built to maintain a relationship rather than complete a task. The harms attributed to them are attachment, disclosure and displacement of human contact, which are not questions of academic integrity. Those harms do not diminish with age in the way that reliance on a homework assistant plausibly does.
One document, two theories. For younger children the argument is developmental and provisional; for companion chatbots it is about the product category itself, with no expiry attached.
What the release does not name
The city's own announcement names no commercial assistant. It names the five permitted pilots and describes the prohibited category in general terms, and it is coverage rather than the district that has reported ChatGPT and Claude as blocked for all grades.
Keep the distinction, because the two lists behave differently. A named-product blocklist has to be maintained; every new entrant is unblocked until someone notices. A category definition applied at procurement covers products that do not exist yet, at the cost of arguments about what counts.
Disclosure: RECATOOLS is written with the assistance of Claude, made by Anthropic, which is named in third-party coverage of this policy.
The screen-time rules travelling alongside
The same announcement sets recommended daily limits of 30 minutes for grades 3 to 5 and 45 minutes for grades 6 to 8, with restricted one-to-one device time for second grade and below. These are not AI rules and they are not framed as temporary.
They may prove the more consequential half. A one-year AI pause affects tools that arrived in the last three years. A daily screen-time ceiling affects the whole one-device-per-child architecture districts built during the pandemic, and it is the first element that would require rewriting existing lesson plans rather than declining new ones.
Michael Mulgrew, president of the United Federation of Teachers, described AI as "a very dangerous thing" and argued schools should hold off on new products until they are properly vetted. That is a procurement position more than a pedagogical one, and it is consistent with a union whose members are the people asked to supervise whatever gets approved.
What a year of evidence can and cannot produce
The stated purpose is to gather evidence. A single school year can deliver only so much, and the coalition will be asked for a recommendation regardless.
A year is enough to establish usage, incident reports, teacher and parent survey data, and whether the five pilots were operationally workable. It is not enough to measure effects on learning outcomes, which need multi-year cohorts and a comparison group. The district has created a natural comparison by permitting pilots for 50,000 high schoolers while restricting everyone else.
Ask in 2027 whether anyone designed the pilots to be compared. If the 50,000 were selected by school interest rather than assigned, the comparison is confounded by exactly the schools most likely to adopt, and the year will produce a decision without producing evidence.
View author profile → · Editorial policy
