> ## Content Index > Fetch the complete content index at: https://www.vitalcitynyc.org/llms.txt > Use this file to discover other available public pages before exploring further. # What a Better Public School AI Policy Would Look Like - URL: https://www.vitalcitynyc.org/nyc-public-school-classroom-ai-policy/ - Published: 2026-09-10T14:55:09.000Z - Updated: 2026-09-10T14:55:09.000Z - Description: New York City’s new rules provide too little room for rigorous experimentation by teachers and students. - Author: Justin Reich - Tags: Education, Technology, Artifical Intelligence New York City’s new policy on AI in schools is based on politics, not science. Still, it’s hard to blame City officials: There is no definitive science of generative AI for learning yet. In March 2026, New York City’s Department of Education [released a Red-Yellow-Green light model](https://www.nytimes.com/2026/03/24/nyregion/ai-nyc-classes-grades.html?ref=vitalcitynyc.org) to guide the use of new generative AI tools in schools. Letting chatbots assign grades, write IEPs for students with disabilities and upload student data were Red, forbidden. Teacher lesson planning and finding operational efficiencies were Green, full speed ahead. Student-facing AI use was Yellow: educators should be cautious, but experimentation was allowed. These initial guidelines were released with little fanfare. The New York Times got an exclusive scoop on the story and printed it on page 17 of the New York edition. Chancellor Kamar Samuels announced the guidelines at a low-key event at a Manhattan high school, befitting the release of a minor piece of technocratic arcana. Last Wednesday, Sept. 2, on the eve of a new school year, [City officials announced a much more restrictive set of guidelines](https://www.nyc.gov/mayors-office/news/2026/09/mayor-mamdani-and-chancellor-samuels-put-students-first-with-nat?ref=vitalcitynyc.org), superseding the guidance from only six months prior. The new policy bans student-facing AI in pre-K-8, sets tight limits on all screen time in pre-K-8 and removes over three dozen existing technology tools with recently added AI features. Students will only be permitted to use generative AI in high school, and their educators can only select from one of five City-approved pilot tools (from writing-assistance apps to vibe coding platforms) — and then only after students complete 90-minute modules about AI literacy. This new policy was announced at a [press conference](https://www.youtube.com/watch?v=P%5FtTStZAM20&ref=vitalcitynyc.org) led by Mayor Zohran Mamdani, with remarks from six City and State elected officials. Councilman Lincoln Restler provided some context for the major change in City policy when he explained, “You know, last year I got dozens of complaints from parents angry that their children were talking into AI programs to learn how to read.” (These parents are probably complaining about AI reading tutors like Amira, which provide AI-generated feedback and adaptively select texts for students to read using automated analysis.) Anyone interested in the politics of AI should watch the press conference. The elected officials do not look like they are wading into a fraught conversation. From their enthusiastic endorsement of the new restrictions, they appear confident that the New York City electorate widely supports restrictions on technology in schools. Nothing in the science of AI and education changed between March and September. There were no new major publications, and no important results showing the harms or benefits of generative AI and learning. There weren’t even any particularly newsworthy events, positive or negative, about AI in New York schools. In the parlance of our time, the vibe shifted. New York City parents, like folks around the country, are increasingly angry with and suspicious of both Big Tech and edtech. [Florida’s statewide school cellphone ban](https://www.the74million.org/article/florida-study-cellphone-bans-promote-academic-gains-after-a-year-or-so/?ref=vitalcitynyc.org), Texas Gov. Greg Abbott’s [about-face on data centers](https://www.nytimes.com/2026/06/10/us/texas-abbott-data-centers-regulation.html?ref=vitalcitynyc.org) and Mamdani’s reversal on AI in schools are part of the same pattern. Parents are worried about the low quality of edtech apps, about recreational screen time replacing learning time and social interaction and about the growing power of AI companies in our economies, our power grids, our politics and our daily lives. One of the magical features of our public schools is that politics always must give way to technocratic operations. Public schools respond to society’s most urgent questions: “How do we live together? What does it mean to be a neighbor, a New Yorker, or an American? How do we raise our children for an unknown future?” They then transmogrify those hopes, fears and longings into bus routes, lesson plans and laptop insurance. The politicians held the floor for one morning, and now the rest of the year belongs to the teachers, students and administrators who will enact, ignore and resist this guidance. I’ve spent the last 25 years teaching with technology, consulting with schools about integrating new technologies, and researching technology and learning at Harvard and MIT. If I were the benevolent dictator of New York City schools, the AI policy would have more opportunities for students to engage with AI tools. I find the new boundaries too restrictive for a technology widely used by young people that is poised to transform the workplace and civic life and sits at the center of this year’s most urgent policy debates. I have fewer quibbles with the bans for the youngest students. The evidence of [edtech’s effectiveness in young grades is limited](https://www.hup.harvard.edu/books/9780674278684?ref=vitalcitynyc.org), and young students have plenty to do learning how to live and work with their peers, develop fundamental skills in reading and mathematics and build habits of mind befitting scholars and citizens. But as students get older, they should have more opportunities to engage with the technologies defining their era. Every New York City high school upperclassman should have an intensive experience using a frontier AI model for academic or creative purposes under the guidance of a caring teacher (even if the point of that investigation is to criticize AI). I’m more enthusiastic about [Technology Civics](https://www.nytimes.com/2026/08/23/business/schools-big-tech-google-microsoft.html?ref=vitalcitynyc.org) than I am about [AI literacy](https://www.teachlabpodcast.com/ai-literacy-part-1-where-angels-fear-to-tread-with-sam-wineburg/?ref=vitalcitynyc.org): Questions of power, politics and policy that will endure in the decades ahead are a better starting point for these new tools than lessons about prompting strategies and AI architecture destined for obsolescence. Opportunities for student leadership should be essential to district policy around new technologies. The greatest city in the world has some of the greatest kids in the world, and young New Yorkers should be leading citywide conversations, co-developing and reviewing plans for the pilots, and gathering feedback from parents, neighbors, politicians and other stakeholders. But I must admit: Despite studying education technology for more than two decades, my position isn’t really based on science either. Nearly four years have elapsed since the arrival of ChatGPT, which seems like a long time, but it’s actually not nearly long enough for education researchers to get a firm handle on new technologies. Google was founded in 1998; the first peer-reviewed research with demonstrably effective techniques for teaching kids how to sort truth from fiction on the internet [was published in 2019](https://eric.ed.gov/?id=EJ1262001&ref=vitalcitynyc.org). If the Big Science of AI in education will take a couple of decades, then schools will have to forge ahead with Local Science. The most important element of the New York City policy is absolutely worth celebrating and replicating across the country: treating our engagement with AI in schools as a systematic experiment. The mayor and the chancellor promised to conduct this year as a trial, collecting data about teaching and learning from classrooms and continuing to gather feedback from students, parents, educators, technologists and community members. New Yorkers should hold the Department of Education accountable to those promises. In the weeks ahead, the City should publish details about the outcomes they plan to measure, the data they plan to collect, the research partners who will help them and their plans for using this year’s findings to revise next year’s policy. For instance, one of the City’s high school pilot tools is called Quill. The city describes it as “designed to support close reading, text analysis and the use of evidence to support claims. Students will use Quill for no more than 15 minutes per week.” To evaluate this pilot, City officials and research partners might start with some basic qualitative assessment. Do students and teachers actually use Quill for 15 minutes a week? (A shocking number of edtech tools that schools pay [for are never used](https://www.the74million.org/article/schools-are-paying-for-ed-tech-that-students-never-use-could-a-new-contract-model-change-that/?ref=vitalcitynyc.org).) Do they wish they could use it more? If you watch students using it, does it look like they are clicking through drudgery, or doing interesting thinking? What are the human-to-human interactions sparked by the tool? Do students and teachers have interesting discussions about the work they do inside the platform? These kinds of anthropological observations can be paired with more quantitative analysis. On standardized assessments of reading and text analysis, do students in Quill classrooms do better than those in similar classrooms not using Quill? Perhaps the most important evidence is to be found in student work and writing. Do any schools piloting Quill have assignments (a Shakespeare analysis, a personal essay) that teachers have consistently used for several years? If so, how will the writing and thinking expressed in student work in 2023 differ from the results after the Quill pilot in 2027? Another pilot tool, Playlab, will be more complicated to evaluate. Playlab lets students and teachers create apps and directly interact with chatbots powered by both frontier models (like Claude and Gemini) and open-weight models (like Llama from Meta or the Mistral models from Europe). It offers some educational scaffolds and data protections that aren’t available from commercial LLM providers, but its main draw is that it provides authentic opportunities to interact with new technologies. Quill should improve student reading and writing, but Playlab could be deployed to almost any educational goal imaginable. Some methods of evaluating learning with Playlab may be similar to approaches to Quill: observing students or asking teachers about their experience. One should not expect much impact on test scores. City officials might consider surveying New York City employers to learn more about the skills they hope their employees show in their use of LLMs, and then evaluate projects to see whether student opportunities are aligned with the local labor market values. Or they might ask City officials and civic leaders to predict how young engaged citizens should use generative AI in the future, and map their responses onto the student projects with Playlab. New York City has the chance to be the most rigorous district in its evaluations, but it certainly won’t be alone in testing new approaches to AI in education. Every other school district in the country is also conducting one kind of experiment or another this year. Los Angeles announced it will be testing [a complete ban of student-facing generative AI](https://www.nytimes.com/2026/09/03/us/lausd-schools-ban-ai-artificial-intelligence.html?ref=vitalcitynyc.org). Gwinnett County in Georgia will continue testing schools designed around [engaging with artificial intelligence ideas at every grade level](https://www.nytimes.com/2026/05/30/opinion/ai-high-school.html?ref=vitalcitynyc.org). Some schools will test an incoherent, unplanned approach. Across the 130,000 schools in the United States, we’ll collectively try a little bit of everything. The challenge for educators across the country is to synthesize the lessons from those experiments as quickly as we can: Which strategies best prepare students to engage with contentious public debates about the role of AI in society? What AI experiences prepare graduates for jobs and civic life? What approaches expose young people to risks of unhealthy relationships with chatbots, privacy invasions or just an unseemly attachment to technology brands at young ages? The best chance of answering those questions is to approach them systematically, as New York has promised: declaring a set of hypotheses and principles, a manageable range of experiments and methods for evaluating and revising our plans. Politics won’t and shouldn’t disappear from school policy. But hopefully in years ahead, AI in education guidance will be increasingly informed by evidence from the kinds of experiments New York City students and teachers will be conducting this year.