New York halts school software purchases: AI in classrooms needs governance before a catalog

By Chalkbeat · July 8, 2026.

New York City public schools Chancellor Kamar Samuels sent an email this week to school principals asking them to hold off on buying new educational software until the Department of Education finishes drafting its final guidance on artificial intelligence, expected by the end of this summer. It is the most concrete step the city has taken so far to rein in the use of AI tools in the classroom, and it comes after months of friction: the initial guidance, published in March, drew a flood of criticism for being seen as too lax, to the point that more than half of the members of the City Council signed a petition calling for an outright moratorium on AI in schools. Samuels himself admitted in May that the district had "missed the mark" with that first version and said the revision would include stricter rules, especially for younger students. The release, promised for June, has been delayed without the Department giving a new date.

The backdrop is as revealing as the pause itself: at a City Council hearing in June, education officials could not say how many schools already use AI products, or which ones. The reason is structural, not anecdotal: software purchases are handled at the level of each individual school, with no centralized registry, a gap that the state comptroller had already flagged as a serious risk. The district has responded by sending a survey to schools to find out, belatedly, what is already coming into the classroom. The freeze excludes software needed for "mandated services" or for the start of the school year, but since many programs —including grade- and attendance-management systems that have been in use for years— require a new purchase order each year, the halt could affect well-established tools, not just AI newcomers. Principals interviewed by Chalkbeat describe the practical problem: budgets already closed, interventions already planned for the school year starting in weeks, and now an added uncertainty about which tools they will actually be able to use.

This story is not so much about AI as about the governance of public procurement in a system that, for years, delegated those decisions to each school with no central visibility. The arrival of generative AI has simply made intolerable an opacity that already existed: if no one knows what software schools use, then no one can know what data on minors those products process, or with what safeguards. In that sense, Samuels's pause is less a brake on innovation than a belated attempt to bring order before adoption, already widespread in practice, becomes irreversible.

Our reading is that this episode is exactly the kind of short-term friction that was to be expected, and that it should not be read as a sign that AI has no place in education. The largest school district in the United States is discovering, several years behind the pace at which these tools have been adopted in classrooms, that protecting children demands a different —and more demanding— standard than the one that suffices for an adult user. That is consistent with what we have already pointed out in other sectors: the gain is not in letting in any tool with an AI label, but in ensuring that whoever deploys it —here, the education system itself— has the capacity to govern it, audit it, and explain what it does with a minor's data. The clumsiness of not even knowing which products are in use is the symptom; the cure is not to reject the technology, but to build the governance infrastructure that should have existed from the start.

In the medium term, this kind of pause and controversy —which will recur in other districts and countries as pressure from families and teachers grows— will probably accelerate the emergence of certification and audit standards specific to educational AI, similar to those that already exist for children's data-protection software. Whoever builds educational AI products designed from the ground up to withstand that scrutiny —data traceability, age controls, transparency about which model lies behind it— will have a real competitive advantage over those who simply added a chatbot to an existing product. And if, as Zendoric's underlying thesis holds, AI ends up freeing teachers from administrative burden to focus on what truly matters —pedagogical judgment, the relationship with the student— that future will only arrive if the adoption process in childhood is carried out with the rigor that New York, late but with evident conviction, is now demanding.