Executive Summary

Key Figures

Sources: NYC Mayor's Office press release (September 2, 2026) and Chalkbeat (September 2, 2026)

Mayor Zohran Kwame Mamdani and Schools Chancellor Kamar H. Samuels announced the policy on September 2 at the Brooklyn STEAM Center, eight days before the school year began. For the 2026-27 school year, students from 2-K, the program that starts at age two, through eighth grade may not use software that runs student-facing generative AI. Companion chatbots are prohibited across every grade. "Children need teachers and human connection in order to learn and grow," Mamdani said in the announcement. Samuels said the city was "standing firmly in our belief that innovation does not mean more technology."

High schools are handled differently. Every high school student has to complete two 45-minute AI literacy modules a year, and five centrally approved pilots are opening: Quill, Edia, Brisk Teaching, Playlab, and Intel AI-Ready Schools. Participation is capped at 50,000 high school students in general education classes, roughly 5% of the public-school student body by the mayor's office's count. Screen time guidance came with the policy. Students in second grade and below are not to use individual devices one-to-one, while the recommended daily caps are 30 minutes in grades 3-5 and 45 minutes in grades 6-8. Assistive technologies, assessments such as screeners and diagnostics, out-of-school learning such as remote instruction days, and centrally approved instruction sit outside those limits.

The ban itself has exceptions too. "Across all grades, this policy does not apply to assistive technology and associated screen time, provided through a student's Individualized Education Plan (IEP) or 504 Plan," the guidance reads, and students with disabilities and English Language Learners keep access to other necessary tools. The mayor's education adviser offered two examples: an eye-tracking technology that helps students with cerebral palsy type, and an Education Department-created program called Hello App that helps students learning English communicate with teachers. In high school, career readiness pathways such as CTE and FutureReadyNYC are carved out as well, where students may use certain AI tools under teacher supervision.

Those are the rules. Actually turning something off comes next. Mamdani said the Education Department is "disabling" AI features in 38 existing citywide ed tech contracts. Where a company cannot simply turn off the AI features of a product, the city will no longer use that product, his education adviser Ailish Brady added. There is a process for the year ahead as well. A new Technology in Schools Coalition of students, educators, parent leaders, elected officials, advocates, union partners, and experts will assess the impact through the school year and deliver a report of policy recommendations in April 2027.

Reactions split. A coalition of advocates that had pushed for a two-year moratorium called the policy a "step in the right direction" while raising concerns about pilot programs that will let tens of thousands of high school students use tools involving generative AI. The criticism ran the other way too. Richard Buery, a former deputy mayor and CEO of the Robin Hood Foundation, said: "While well-intentioned, a blanket one-year AI ban, without exceptions, on all uses through middle school is not the right call." His reasoning was that educators already use tech tools to quickly generate lessons parents can use at home or to get faster feedback on classroom assignments. Through middle school, the only exception written into the AI ban itself is assistive technology.

The number arrived without its contents. Officials haven't provided a list of those companies, Chalkbeat reported. A larger gap sits beside it. City officials have admitted they have little hard data on how much AI is currently in use across the 1,600 schools. A survey did go out asking schools about AI use, and officials declined to release the results despite repeated requests.

Only two products have been named. Brady said the widely used digital tutor Amira and the digital version of the Houghton Mifflin Harcourt reading curriculum will both "turn off" their AI components. The HMH curriculum is the most popular of the three mandated NYC Reads literacy programs, Chalkbeat noted. HMH answered differently. A spokesperson said "there is no student-facing AI in Into Reading," the company's elementary school reading curriculum. An Amira spokesperson said the product is currently used in 222 schools and that the company has reached out to the Education Department for clarity on what the new policy will mean for it. Google's chatbot Gemini comes pre-loaded on many students' Chromebooks, and Brady didn't directly respond to a question about whether it would be disabled, though advocates who met with city officials said they were told it would be shut off.

The 38 has a defined perimeter. New York City principals buy education technology in two ways. For smaller purchases they can contract directly with a vendor, and for larger ones they generally have to use a vendor that already holds a citywide contract with the Education Department. The 38 sit inside that second category. Whatever a school bought on its own falls outside the count.

Step outside the class schedule and the line shifts again. Any instruction by community-based organizations during school hours, in NYCPS buildings or contracted settings, must abide by the department's AI and screen time policies. Afterschool programs run under the oversight of the Department of Youth and Community Development or are operated by those organizations rather than the district, and there the guidance says NYCPS will provide policy aligned guidance to providers. The tool the same child uses in the same building at 4 p.m. moves into that softer territory.

Filling the gap has already begun to cost something. After freezing all new education purchases for nearly two months, officials let principals resume buying software on the day of the announcement, with a condition attached. Any new request for ed tech under $25,000 now goes through an extra layer of review from the Education Department's IT, legal, and contracts teams. Deputy Chancellor Lindsey Oates warned principals that the reviews will "take some time," adding a bottleneck to a contracting process that could already be slow. The first thing the ban created for the department that declared it was a need for more review capacity.

The policy document the Education Department published for families names the process. "Before any tool can be used with your child, it must be reviewed and approved. We call that review ERMA—the Enterprise Request Management Application." Under New York State Education Law 2-d and its regulations, every vendor the district works with must sign an agreement about how it will handle student information, provide NYCPS an explanation of what data it collects and how it protects it, and meet recognized security standards. ERMA, the document says, is where the district confirms all of that.

So a vendor disclosure register already exists. What gets written into it is data handling. Whether generative AI sits inside the software, and whether that capability reaches students directly, is not among the fields a vendor files. The department drew that boundary itself: "The ERMA process currently reviews tools for data privacy and security. NYCPS is committed to building expanded evaluation capacity, including areas not covered by ERMA such as algorithmic bias, equity impact, and instructional effectiveness." That sentence sits in the column of things promised for later.

The same boundary shows up in what the department did approve. It listed how it picked the five high school pilots, and the first of four criteria is "compliance approval with the NYCPS Enterprise Request Management Application (ERMA) data privacy and security review process." The other three are judgments: tools that support writing or math where the student does their own thinking, applications that teach teachers to build their own student-facing tools for their students' specific needs, and tools that directly enable project-based learning. Only the first can be answered by checking whether an existing process was cleared. The rest have to be argued case by case.

That is why counting the 38 looks like a one-off project rather than a standing procedure. The mayor's office wrote that under the new policy all technology tools used in NYCPS will undergo an exhaustive review and will be barred if their functionality is ultimately determined not to be mission-critical to learning. The review is to include vendor accountability for safety and transparency standards, ethics, evidence of learning design and instructional impact, prior research, and the continuous collection of user feedback. Broad as a principle, but not a list of fields a vendor is required to file.

An exhaustive review is a stocktake. A stocktake is the raw material of an asset register, not the register itself. Once you have counted, you need a way to keep the count true, and keeping it true means vendors have to say what changed each time they change the product. The software industry has spent the past several years working the same shape of problem in security under the name bill of materials. The entries change from vulnerabilities to generative AI components. No reporting has said New York City has mandated any such disclosure. What the announcement suggests is that trying to enforce a ban put the city in need of one.

The rules that apply to teachers sort by purpose rather than by product. Teachers and other school staff may continue using AI for instructional planning and operational tasks, including translating materials and drafting communications to families, subject to the existing ERMA approval process and district procurement rules. On the other side of the line, the guidance states that "AI may not be used for grading, behavior monitoring, or for placement, promotion, graduation, and other decisions about students." It adds separately that all special education documents are developed by qualified professionals, and that counseling, crisis intervention, and therapeutic support are provided only by qualified staff members.

The line is drawn; the machinery for holding it is not. Officials didn't spell out any clear enforcement mechanism for ensuring schools adhere to the AI policy, Chalkbeat reported. Mamdani conceded much of the enforcement will have to happen at the classroom and school level. "Whenever there's been a shift, teachers have had to be at the forefront," he said at the press conference. "And so we trust that teachers, principals, superintendents will be there." United Federation of Teachers President Michael Mulgrew expressed concern that it will largely fall on "school communities or individual educators to figure out after the fact if what their school uses fits the new policy."

Here the list problem from Section 2 meets the purpose problem in this one. Lists are built per product; the rules are attached per purpose. Telling teachers not to use AI for grading does not point at grading tools, it points at a use. The same chatbot that helped plan a lesson can score answers in the next window, and both uses happen through one account on one screen. A product inventory cannot tell them apart. Telling them apart requires a record not of what is installed but of what it was used for.

That kind of record is designed into the pilots. Schools apply case by case and each application is reviewed individually, the mayor's office caps participation at five classes per high school, and the department wrote that "we will ensure that students are participating in only one pilot." Expected usage per tool is tabulated as well: Quill 15 minutes per week, Edia 20 minutes, Brisk Teaching 20 to 40 minutes, Playlab 45 minutes with two assignments per term, and Intel AI-Ready Schools one period per week. The population where who used what and how much gets written down this way is the 50,000 in the pilots. There is no equivalent table for the 600,000 covered by the ban.

For companies this is a familiar ask. Recording which system processed which data for which purpose has been handled for years under the name data lineage. That is not to say schools have already gone that way. The opposite is what the record shows: the school year opened with rules drawn by purpose and no method yet for capturing purpose.

It is hard to fault New York City for not having done this already. An organization of 1,600 schools set a policy and had eight days before it took effect. What the case leaves behind is a question of order. An institution that had positioned itself as a cautious adopter found that the moment it decided to block adoption was the moment it had to count what it was already running. A policy can be announced in a day. Asset visibility cannot be built in one.

Companies sit in the same structure. Which of the SaaS products deployed inside a firm have generative AI in them is not something rereading the contracts will answer, because contracts mostly govern data handling and liability rather than product components. What New York City's ERMA confirms occupies exactly that position. The three questions below are not reported findings; they are what an organization facing the same problem might reasonably put to itself.

• Can you say today how many of the software products you run contain generative AI? If you cannot, what is missing is the list, not the AI.
• Is that determination filed by the supplier, or investigated by you each time? Investigation answers the question once, and cannot keep pace with product updates.
• If your rules differ by purpose, where does the purpose get recorded? Not in a product register.

Official Documents

• 1.New York City Public Schools. (2026). "Guidance on Artificial Intelligence (AI) and Screen Time." Policy text. Source for the ERMA and Education Law 2-d passages and the pilot selection criteria in Section 3, the teacher use rules and the per-tool dosage table in Section 4, the grade-band table and exceptions in Section 1, and the afterschool and community-based program scope in Section 2.
• 2.Office of the Mayor, City of New York. (2026-09-02). "Mayor Mamdani and Chancellor Samuels Put Students First with Nation's Broadest Generative AI Moratorium in Schools." Source for the 600,000 students and two-thirds figure, the companion chatbot prohibition across all grades, the five pilots and the 50,000 cap for general education classes, the five-classes-per-high-school limit, and the exhaustive review language.

News Coverage

• 3.Elsen-Rooney, M., Zimmerman, A. (2026-09-02). "NYC's new school AI ban explained: What's changing? What's still murky?" Chalkbeat New York. Source for the 38 contracts and the missing list, the absence of hard data across 1,600 schools and the withheld survey, the Amira, HMH, and Gemini reporting, the exception examples of eye-tracking technology and Hello App, the $25,000 review, the advocacy coalition and Richard Buery reactions, and the Mulgrew comment.
• 4.Elsen-Rooney, M., Zimmerman, A. (2026-09-02). "NYC to ban student AI tools in 2-K through 8th grade and limit classroom screen time." Chalkbeat New York. Same-day first report.
• 5.Lim, D. (2026-09). "New York City public schools ban AI." AI Times. Korean-language coverage, used for Korean transliterations of names in the Korean edition of this article.
• 6.ABC7NY. (2026-09). "New York City public schools banning AI use through middle school this year." Local broadcast coverage.

CNN's same-day story (cnn.com/2026/09/02/tech/new-york-city-classroom-ai-ban) was blocked to us, so its text could not be verified and this article does not cite it.