The School District AI Ban in New York City and Los Angeles: What the Exemptions Tell District Leaders

Estimated reading time: 6 minutes

Within two days of each other, the two largest school systems in the United States told their students to stop using generative AI. The school district AI ban in New York City covers roughly 600,000 children through eighth grade; the one in Los Angeles covers every student on a district device. The headlines read as a verdict on AI in education. The documents underneath read as something narrower and more useful: a decision about which tools have a defined purpose and which do not. New York names the students it will not cut off. Los Angeles, so far, does not.

What the two districts actually decided

New York went further than the grade band suggests. Companion chatbots are prohibited at every grade with no pilot and no supervised path, and general assistant services are blocked district-wide. Around 38 existing ed-tech products had their generative features disabled or stripped out. High schools get two 45-minute AI literacy modules a year and a tightly capped pilot: five vetted tools, no more than five classes per school, reaching at most 50,000 students. Quill is limited to 15 minutes a week, Edia to 20.

Los Angeles did the opposite of granular. The block is enforced at the device, applies to every age, and was put in place administratively. Board member Nick Melvoin’s reaction at the meeting was not about the merits: “If the policy is we’re not going to allow students to use generative AI for the meantime, then we should publicize that.”

The exemptions are the story

The New York policy exempts five things by name. Two of them matter enormously to anyone running a language programme:

• “Assistive technology for students with disabilities”
• “Multilingual learners”
• Career-readiness programmes such as computer science
• Centrally approved instructional programmes at every grade, including e-books and coding software
• Some student assessments

Read carefully, the nation’s broadest moratorium on classroom AI is not a judgement that the technology fails children. It is a judgement that open-ended, general-purpose chat has no defined instructional job in an elementary classroom, while a centrally approved tool with a named population and a bounded use does. The same document that removes generative features from 38 products also protects the ones a multilingual learner needs to reach the lesson.

Teacher provisions carry the same logic. New York teachers may use AI to plan instruction, translate materials and handle operational work. They may not use it to grade assignments, to make promotion or graduation decisions, or to counsel a student in crisis. The line is drawn at consequence, not at technology.

What Los Angeles left open

Los Angeles enrolled 497,105 students in 2025-26, of whom 81,328 were English learners, 16.4 percent of the district. A block enforced at the device reaches all of them, because a device-level filter does not know which student is holding the Chromebook or why.

That is not an argument that Los Angeles was wrong to act quickly. It is an observation that a filter is a blunt instrument and an exemption list is a precise one, and that Los Angeles has adopted the first without yet writing the second. The ad hoc committee now has that work in front of it. Districts drafting their own policy this autumn should notice which of the two models they are copying, because the choice determines whether a newcomer keeps access to the tool that lets her follow her science class. Our guides to English Learners in New York and English Learners in California cover the programme context in each state.

This is not a reversal of the federal position

It would be easy to read these two decisions as districts breaking with Washington. They are closer to an aggressive application of it. Three weeks earlier, the U.S. Department of Education’s Dear Colleague Letter on education technology told districts to stop arguing about screen minutes and to judge every tool by whether it demonstrably improves learning, and to remove the ones that cannot show it.

Thirty-eight products losing their generative features in a single district is what that instruction looks like when a system actually carries it out. The federal letter asked what learning problem a tool solves, for whom, for how long, and on what evidence. A general-purpose chatbot in a third-grade classroom answers none of those four. A translation and comprehension support named in a student’s language plan answers all of them.

What district leaders should do now

1. Separate your list before your board does it for you. Sort every licensed tool into general-purpose chat, embedded generative features inside instructional software, and purpose-bound tools tied to a named population. The three deserve different answers, and a policy written in a hurry will treat them the same.
2. Write the exemptions before the block. Assistive technology and language access are the two categories that produce compliance exposure when they disappear. Naming them takes an afternoon. Restoring access after a filter has been running for a term takes longer.
3. Decide who signs. Los Angeles shows what happens when staff act ahead of the board. Whatever your district concludes, the decision should be published where families and principals can find it.
4. Keep the evidence question. A moratorium buys a year. It does not answer what a tool was supposed to do, and the review at the end of the year will ask. See our guide to how districts buy ed-tech for where that question belongs in a contract.

Questions district leaders should ask

• If we blocked generative AI at the device tomorrow, which supports would our multilingual learners and students with IEPs lose that afternoon?
• How many of our licensed tools have generative features we did not specifically buy?
• Who in our district can produce the list of AI-enabled products, and how current is it?
• Does our language access plan name any tool that a district-wide filter would catch?
• If our board asked for a policy in two weeks, whose draft would we be adapting?

Frequently asked questions

Is the school district AI ban permanent?

No. New York’s moratorium runs one school year, after which a body called the Technology in Schools Coalition is to publish a report with recommendations. Los Angeles described its block as lasting the school year and until further notice, with a board committee developing recommendations in the meantime.

Does the New York policy apply to multilingual learners?

Multilingual learners are one of five named exemptions, alongside assistive technology for students with disabilities, career-readiness programmes, centrally approved instructional programmes and some assessments. The moratorium targets student-facing general-purpose generative AI, not every tool with an algorithm in it.

Can teachers still use AI in these districts?

Yes in both. New York permits teacher use for instructional planning, translating material and operational tasks, while explicitly prohibiting AI for grading, for promotion or graduation decisions, and for counselling a student in crisis. Los Angeles did not bar teachers from using AI on district devices.

Are other districts following?

Some are moving. Roughly half of the candidates for Chicago’s Board of Education signed a pledge supporting a three-year classroom moratorium, and districts in Colorado, Kansas, Minnesota, Texas and Washington have blocked or limited tools such as ChatGPT, several of them allowing approved platforms under teacher supervision.

The takeaway

Two districts reached the same headline through different instruments. One wrote a list of who keeps access and why; the other switched off a category and will decide the exceptions later. For a district leader drafting a policy this term, that difference is the whole lesson, and it will be felt first by the students who were relying on the tool to understand the lesson at all.

Related reading

• Education Technology Guidance: The Five Questions ED Asks
• Apps for English Language Learners: A K-12 District Guide
• AI in Education: Guides for District Leaders
• Complete Guide to English Learners in U.S. Schools

Sources: Office of the Mayor of New York City, September 2, 2026; BetaNYC policy analysis; Education Week on New York; Education Week on Los Angeles; Stateline; California Department of Education district profile.

The Students an Exemption Is Written For

A newcomer arrives in October. She has a science class, a teacher with twenty-eight other students in the room, and no way yet to say what she understood. For the first few weeks she says almost nothing at all, and everyone involved knows that the silence is not comprehension.

What she needs is not complicated. She needs to talk, often, with something patient enough to wait for the sentence to arrive. What makes it hard is arithmetic rather than skill: a teacher splitting one class period across a full roster cannot give any single child more than a couple of minutes, and a newcomer needs far more than that to move.

This is why the districts writing careful policy are naming language access as an exemption rather than a loophole. The tools that earn a place on that list are the bounded ones: a named group of students, a defined dose, and reporting a coordinator can put in front of a board. Telo AI runs five-minute sessions three times a week, reports participation and speaking growth on a WIDA-aligned view, and keeps none of the conversations themselves: no audio, no transcripts, and no student data used to train AI models.

See what a five-minute speaking session looks like for a newcomer

A district that can say which students a tool serves, and show what changed for them, is not the kind of district that has to switch anything off in a hurry.