Early September brought the clearest policy statements yet from the two largest school districts in the US. On September 2, New York City announced a one-year moratorium on generative AI for students from 2-K through eighth grade, covering nearly 600,000 children. Companion chatbots are prohibited in every grade, and all classroom technology will face a new review that excludes anything not essential to learning. The same week, Los Angeles Unified suspended student access to generative AI on district devices across all grades while a committee drafts guardrails.
What the districts actually did
It is worth reading both decisions carefully, because they are narrower than the headlines suggested. New York's policy is a pause with carve-outs: up to 50,000 high school students will use five vetted tools in teacher-supervised pilots, every high schooler gets AI literacy modules twice a year, teachers keep AI for planning, and assistive technology for students with disabilities is protected. Los Angeles acted through administrators rather than a board vote, which surprised some board members. Neither district is saying AI has no place in education. Both are saying that no one has yet shown them it belongs there on the current terms.
That instinct is reasonable. Districts were handed products faster than evidence, and the summer's news about procurement scandals and contract investigations gave officials every reason to slow down. A moratorium buys time to ask who built a tool, what it collects, and whether it works.
Why a pause is not an answer
But a moratorium governs access, and access was never the whole problem. A student who is barred from a school-issued chatbot still has a phone. The harder question, the one this series keeps returning to, is what a school can know about what a student actually understands when the work is done out of sight.
Banning a tool does not restore the evidence that unsupervised work used to provide. Neither does approving one.
We want to be careful about what we are not claiming. These policies are not mistakes, and they are not a verdict on AI in learning. Pausing is a legitimate response to uncertainty, and the districts built in review and pilots that should generate real data. The risk is subtler: that a pause gets read as a solution, and the assessment question waits behind it. A year is a long time to postpone deciding how students will show their thinking.
What to ask during the pause
The more useful reading of September is as a window. Both districts have said they will use this period to evaluate tools and write guardrails. The questions worth putting into that process are practical ones. What counts as evidence of learning when a tool is allowed, and when it isn't? Where does a student explain their own reasoning, in their own words, in real time? And who reviews the results, so that a flag from any system is something a teacher examines rather than a verdict a student inherits?
If the pilots in New York report honestly, whichever way they point, we will learn something no policy statement can tell us. Until then, the oldest test in teaching still applies, and no moratorium changes it: can you walk me through what you did, and why?