New York City has taken one of the most forceful steps yet by an American school district against generative artificial intelligence. Mayor Zohran Mamdani announced a one-year moratorium on AI tools for students from early education through eighth grade. The policy will affect roughly 600,000 children when it begins with the 2026-27 school year. The directive removes all student-facing generative AI software from classrooms and explicitly bans companion chatbots outright.
Key facts at a glance
- The one-year moratorium covers approximately 600,000 students in New York City public schools from pre-K through eighth grade.
- It begins in the 2026-27 academic year.
- All student-facing generative AI software will be removed, and companion chatbots are banned outright.
- High schools will not face the same ban, but only five pilot programmes can operate, limited to five classes per school.
- Norway banned generative AI for grades one through seven in June; the restriction took effect at the end of August.
- EU law currently classifies some school AI systems as high risk but does not directly ban AI tools used by children.
A deliberate refusal
Mayor Mamdani framed the decision not as a temporary technical adjustment but as a refusal of the technology industry’s preferred future. “The tech industry wants us to believe that AI-powered early education is not only inevitable, but necessary,” he said. “We do not see it that way.”
The word “refusal” matters. Many school districts have approached generative AI by issuing guidelines to teachers and principals. New York is instead removing tools from the classroom environment. The mayor’s office said all student-facing generative AI is being eliminated. Companion chatbots, which are often designed with conversational personalities and may encourage long conversations with children, are banned because of the risks they pose. This is a stronger intervention than a typical “wait and see” recommendation, and it signals that city officials see certain AI tools as a potential threat to learning during the developmental years.
What the moratorium means for schools
The ban does not apply equally across all grades. Education officials in New York have focused on the period from early-childhood programmes through eighth grade, when students are building foundational reading, writing, and reasoning skills. During these years, educators worry that generative AI can bypass important cognitive work. A child who chooses to ask a chatbot for an answer may not practice the skills that a teacher wants to assess. The ban attempts to keep those tools away from young students for one full academic year. It is a designed pause, not a permanent prohibition.
The city will, however, allow high schools to experiment under close supervision. Up to 50,000 high-school students can join five pilot programmes. Each pilot school can only use AI in five classes. Twice a year, high-school students outside the pilot programmes are expected to complete literacy modules about AI. Those modules are intended to give older students a base of knowledge about how the technology works, what it can and cannot do, and how to use it responsibly.
Why high schools are treated differently
The different treatment of high schools reflects how AI has become embedded in upper-level education. High-school students may need to understand generative AI before entering university or the workforce. They are also more likely to have developed the critical judgement needed to question a machine’s output. In contrast, an eight-year-old may treat a chatbot as a trusted friend. City policymakers appear to believe that a pilot model offers a compromise: schools can test AI without turning every classroom into an unregulated experiment.
By setting the pilot cap at five classes per school, New York is trying to keep the initiative small enough to be evaluated. Teachers in pilot schools will probably collect evidence about how students interact with AI, whether their writing improves, and whether the tools encourage independent thought. The twice-yearly literacy modules for everyone else also reinforce that the city does not want students to be passive consumers of AI, even if they are not using it in class.
A European precedent in Norway
The New York ban is not entirely new territory. Norway reached a similar position three months earlier. In June, the Norwegian government announced a ban on generative AI in primary schools for grades one through seven. The ban took effect at the end of August. It also limits students under 16 to supervised use of AI. Norway’s Prime Minister Jonas Gahr Store said that AI increases the risk that children will skip important steps in their education. His comments point to a concern shared by many teachers: when a machine can answer questions instantly, students may avoid the struggle that produces genuine learning.
Norway’s decision is significant because it was made at the national level. In the United States, education is primarily a state and local responsibility, which is why a city like New York can create a rule that applies to hundreds of thousands of pupils. Norway’s nationwide ban, by contrast, covers all public primary schools and sends a clear signal to technology companies about the limits of the Nordic education market.
The EU AI Act has a gap
In the European Union, the situation is more complicated. The EU’s AI Act classifies educational uses of AI as high-risk under Annex III. That category covers admission decisions, evaluation of learning outcomes, assessment of the appropriate level of education, and monitoring during tests. Each of these is an AI system applied to a pupil by an institution. The regulation covers the system that admits a child into a school and the software that detects prohibited behaviour in an exam hall. It does not cover the chatbot that helps a student write an essay at home.
This distinction matters. The EU has tried to regulate AI from a risk-based perspective, but its rules are written around decision-making and assessment. A child sitting alone with a chatbot is not making a decision under the same framework. That leaves a large area of activity outside the direct scope of the AI Act. Education officials who want to limit the use of AI by children may have to look at national laws or local policies instead.
There is also a timing problem. The obligations for high-risk education systems are not in force yet. The European Commission’s Digital Omnibus proposal pushed the compliance deadline for Annex III applications to 2 December 2027. Until that date arrives, the AI Act will not fully regulate many of the educational systems that were originally intended to be covered.
Italy and the AI literacy exception
Some European countries have not waited for Brussels to move. Italy’s Law 132/2025 has required parental consent for children under the age of 14 to use AI since October 2025. That makes the 2023 Italian experience, when regulators temporarily banned ChatGPT, part of a broader cautionary pattern. Italy was among the first Western authorities to force a major AI product offline because of privacy concerns.
The only EU-wide obligation that is already binding is Article 4, which requires deployers to make sure people have adequate AI literacy. Since February 2025, this duty has applied in all member states. That may seem modest, but it creates a requirement that every school, hospital, and public institution using AI must train people in basic understanding of the technology. New York’s decision to include twice-yearly literacy modules for high-school students mirrors this European approach. The EU wrote AI literacy into law before it could agree on what a ban on children’s use of AI would look like.
A pattern of local action
Since 2023, when Italy set a defining precedent at the national level, the pattern has been consistent: education ministries and national parliaments are setting limits on children’s AI use