More than 1.1 million students in the nation’s two largest school districts are beginning the 2026–27 school year under new limits on generative artificial intelligence, after New York City imposed a one-year moratorium for students through eighth grade and Los Angeles Unified disclosed a districtwide moratorium on student use of generative AI on district-owned devices. The policies are not identical, but together they mark the clearest large-system retreat yet from the assumption that student-facing AI should simply become another classroom technology.
New York’s rules affect nearly 600,000 students in 2-K through grade 8, roughly two-thirds of the city system’s enrollment. The district’s city policy bans student-facing generative AI in those grades, while allowing limited, supervised use in high school. Los Angeles Unified, which enrolls more than 520,000 students according to the district, has gone further on school-owned devices by placing all student generative-AI access under a moratorium while it develops a longer-term framework.
The shift matters because it is occurring at the same moment AI is becoming easier for schools to adopt. Google expanded Gemini in Classroom to K-12 students of all ages in August when administrators have enabled access, according to a Google update. Federal policy is moving in the opposite direction from outright restriction: the U.S. Department of Education’s final AI priority encourages responsible AI use, AI literacy and educator development in discretionary grant programs. The result is no longer a simple debate over whether schools will use AI. It is a governance question about which uses are educationally justified, at what ages, under whose supervision and with what evidence.
New York Separates AI Literacy From AI Access
New York City’s policy draws a sharp developmental line. Students in 2-K through eighth grade may not use student-facing generative AI, while high school students must complete two 45-minute AI-literacy modules covering how AI works, privacy, bias and appropriate use. The system will also permit five centrally approved high-school pilots—Quill, Edia, Brisk Teaching, Playlab and Intel AI-Ready Schools—with limited exposure and teacher supervision. Assistive technology required for students with disabilities and tools needed by multilingual learners remain available.
The city is also disabling the generative-AI components of more than 38 previously allowed educational products that do not meet its new standards. Mayor Zohran Mamdani said the one-year policy is intended to preserve human interaction and what educators often call productive struggle: the effort required to reason through a problem instead of immediately outsourcing it. Reuters reported that the city described the move as the most restrictive school-district AI policy in the country to date.
That does not make New York an anti-AI system. Teachers may continue using approved AI for lesson planning and administrative work, and high school students will encounter the technology in structured settings. The policy is instead an attempt to separate learning about AI from unrestricted learning through AI. That distinction is likely to become more important as general-purpose chatbots are embedded into tools students already use rather than introduced as stand-alone products.
Los Angeles Adds a Second Large-System Test
Los Angeles Unified’s moratorium surfaced publicly at the September 2 meeting of the school board’s AI committee. District administrators said generative AI had been disabled for students on district devices for the school year, a change that surprised some board members because it had not been adopted through a new board vote. The Los Angeles Times reported that the moratorium does not yet govern what students do on personal devices, making the policy as much an infrastructure control as a comprehensive behavioral rule.
Los Angeles is also moving under a broader screen-time strategy. Earlier this year, the district adopted limits that eliminate student screen use before second grade and phase in age-based limits for older students. The AI moratorium therefore fits into a larger reconsideration of when digital tools add instructional value and when they merely add exposure. The distinction matters because “AI use” can describe very different activities: a chatbot writing an essay, an adaptive system changing question difficulty, speech recognition supporting a learner with a disability, or a teacher using a model to draft a lesson.
The Evidence Supports Caution, Not a Simple Ban-or-Adopt Answer
National data show how quickly policy has lagged behind adoption. In the federal School Pulse Panel, only 31% of public schools reported a written student-AI policy in December 2024, while 42% of school leaders viewed student use of AI for education favorably and 69% viewed teacher use favorably. Just 14% said all students were being taught about ethical or appropriate AI use, according to NCES. Those figures predate today’s school-integrated AI tools, but they show that access was spreading faster than consistent instructional rules.
The learning evidence is more nuanced than either boosters or critics often imply. The OECD concluded in its 2026 Digital Education Outlook that generative AI can support learning when it is designed around clear pedagogical goals, but that general-purpose tools can improve task performance without producing durable learning gains. The organization warns that offloading cognitive work can reduce the effort required to build knowledge, even when the immediate assignment looks better.
Recent youth-use data reinforce why districts are acting before the research is complete. A nationally representative August survey from Common Sense found that 70% of U.S. teens reported using AI for schoolwork, and nearly two-thirds of AI-using teens described practices that the researchers classified as offloading critical thinking. The American Psychological Association’s September review of the emerging literature similarly emphasized that the science remains early; one highlighted study found students could perform better with ChatGPT assistance during practice but worse after the tool was removed, underscoring the difference between completing work and acquiring skill, according to the APA.
Those findings do not establish that AI should be excluded from classrooms. They do establish that districts need to specify the outcome they are trying to improve. If the goal is faster drafting or personalized practice, AI may show an immediate advantage. If the goal is durable recall, independent reasoning, writing fluency, mathematical understanding or social learning, the relevant evidence is different and often thinner. That is why New York’s limited pilots may ultimately be more informative than either a permanent ban or unrestricted access: they can be evaluated against defined learning goals rather than adoption rates.
Procurement Is Becoming Education Policy
The most consequential part of the New York and Los Angeles moves may be institutional rather than philosophical. Both systems are asserting that vendors do not get to determine classroom adoption simply by embedding AI into products districts already license. That is a significant shift as technology companies move AI from optional chatbots into learning-management systems, search, writing tools and productivity suites.
For districts, this raises practical questions that conventional software reviews often did not need to answer. Administrators now need to know whether a feature generates student work, merely provides feedback, stores prompts, trains on student data, can be disabled by age or school, exposes students to open-ended conversation, and allows teachers to see how it was used. They also need evidence that the tool improves the learning outcome it claims to support; privacy compliance alone is not evidence of instructional effectiveness.
The policy divergence is likely to remain substantial. Federal education policy currently encourages AI literacy and responsible adoption, while large districts are imposing local restrictions and states are developing their own rules. That decentralization means school leaders cannot wait for one national standard to settle the question. They will need their own definitions of acceptable use, age appropriateness, teacher supervision, assessment integrity and evidence thresholds.
The Educator's Takeaway
For superintendents, principals, teachers and governing boards, the immediate lesson is not that New York or Los Angeles has found the final answer. It is that AI policy now has to distinguish among developmental stage, instructional purpose and type of tool. A district can restrict open-ended student chatbots in younger grades while still teaching AI literacy, permitting assistive technologies, allowing carefully designed high-school pilots and using AI for teacher workflow. Treating all of those activities as one category obscures the real decisions.
For districts evaluating student-facing AI, the most decision-relevant questions increasingly center on the educational objective being served and the human-oversight model surrounding its use. That makes several variables especially important: what students are expected to learn, what data the system receives, what teachers can observe, what happens when the tool is removed and whether the same outcome can be achieved with less technological exposure. New York’s one-year moratorium and controlled pilots, and Los Angeles’s device-level pause, will be meaningful only if they produce evidence that can guide what comes next. The central question is shifting from whether schools are “for” or “against” AI to whether a particular use demonstrably improves learning without displacing the human work required to achieve it.
