New York bans generative AI for under-13s but lets teachers use it: the education paradox

By Delfino.cr · September 4, 2026.

New York announced this week a one-year moratorium on the use of generative artificial intelligence —systems capable of producing text, images or code from a prompt, such as ChatGPT or Gemini— among students in its public schools through eighth grade. The restriction covers some 600,000 students, according to the opinion piece published in Delfino.cr, and is paired with limits on screen time. High school students are not cut off from the technology: for them there will be "limited, supervised experiences" and AI literacy modules.

Mayor Zohran Mamdani summed up the political rationale in a line that is hard to argue with: children need to learn to think for themselves. Columnist Estíbaliz Pérez Pérez, a specialist in innovation and public policy in education, shares the goal but questions the means: is removing a tool from the classroom enough to teach a student to think? Her thesis, which we share, is that it is not, because the problem is considerably older than ChatGPT.

There is a revealing detail in the fine print of New York's policy that the author herself underlines: while generative AI is banned for students, teachers are still allowed to use it to plan their classes and for other professional tasks. The rule does not distrust the technology itself; it distrusts handing it to those who do not yet have the cognitive foundations to use it with judgment of their own. As the article describes, the policy also includes privacy controls, tool evaluation, accessibility exceptions and supervised pilots: this is not a blind ban, it is a managed one.

The underlying argument matters more than the specific measure. For decades, much of schooling has asked students to memorize facts to repeat them on an exam, to complete questionnaires whose answers are literally in the text, or to hand in summaries no one will later discuss. None of that needed artificial intelligence to produce superficial learning: the absence of the tool never generated critical thinking by spontaneous generation. As the author illustrates with two hypothetical classrooms, the difference between a student who copies a chatbot's answer and one who uses it as a starting point to cross-check sources, spot errors and defend a position of their own does not lie in the technology. It lies in the design of the activity, what the author calls pedagogical mediation.

That distinction carries an uncomfortable consequence for any policy limited to permitting or prohibiting: it shifts the problem toward something far harder to solve by decree, namely teacher training. Banning a tool is a one-day decision. Training an entire teaching corps to design cognitively demanding activities, redesign assessment and teach students to verify answers that "sound right" without being correct is a project spanning years and sustained budgets. That is probably the real bottleneck for any school AI policy, well beyond New York.

There is also an equity argument worth keeping in view. Banning AI at school does not remove it from students' lives: those with devices, connectivity and family support at home will keep experimenting with it outside the classroom; those without will have their first contact later and with less guidance. If public schools give up on teaching how to use these tools with judgment, they delegate that literacy to each family's socioeconomic background, precisely the opposite of the leveling role they are supposed to play.

Our take at Zendoric: this miniature debate about a New York classroom foreshadows the larger debate we already have on our hands over AI and employment. In both cases the short-term temptation is to contain what feels uncomfortable —block a tool, slow down a technology— when the real problem is institutional: what judgment we cultivate, which tasks we stop valuing because a machine already does them in seconds, and who ends up better or worse prepared for the transition. In the labor market we have already seen that routine, administrative work is the first to erode, while expert judgment and human interaction hold up; in school, rote memorization is the "administrative work" that AI renders irrelevant, and critical thinking —questioning, cross-checking, arguing— is the equivalent of the judgment that must be deliberately protected and cultivated.

In the long run, if this transition is managed well, a generation that learns to live alongside systems capable of summarizing, drafting or solving in seconds need not come out worse prepared than previous ones: it may come out better prepared, with more free cognitive time for what was barely taught before —thinking, creating, deciding with judgment of one's own. It is the same promise of abundance we defend for the labor market, applied to the classroom: the machine absorbs the mechanical task, the person keeps what genuinely requires judgment. But that promise does not fulfill itself, nor does it arrive by decree: it demands sustained investment in teacher training, something that rarely accompanies the flashiest announcements.

Costa Rica, which as the author argues will have to make its own decisions about AI in the classroom, would do well to look at New York not as a model to copy or discard, but as a regulatory experiment underway from which there is already something to learn: the question that matters is not whether to permit or prohibit, but when, for what purpose and with how much teacher preparation a tool like this enters the classroom.