New York City reversed its ChatGPT ban within five months

The classroom note

On 3 January 2023, a New York City Department of Education spokesperson told reporters that ChatGPT would be restricted on the department's networks and devices, citing 'concerns about negative impacts on student learning' and worry that the tool 'does not build critical-thinking and problem-solving skills'. Five months later, on 18 May 2023, Chancellor David Banks published a first-person account in Chalkbeat's education-leader section reversing course. He wrote that the earlier response had 'signaled fear and risk' and overlooked what generative AI could do for students and teachers who would work in a world shaped by it. For a Queens classroom, the practical change was concrete: a tool invisible on school devices in January became something teachers were told to explore by May, with the department promising a toolkit and shared examples rather than a rule.

What the evidence says

Both statements are the district's own words, not outside reporting about the district. The January statement, attributed to spokesperson Jenna Lyle, gives a reason grounded in learning and accuracy concerns rather than a technical problem. The May reversal, under Banks's own byline, gives different evidence: an admission the first decision was made quickly, without weighing instructional use. Neither document offers a study, a pilot result or a measured outcome from the intervening months; the reversal describes 'educators already teaching about AI ethics' and using the tool for lesson planning, but names no source, sample or period for that claim. Read together, the two texts document a reversal in stated reasoning, not one supported by new measurement.

The implementation question

The mechanism behind both decisions was access control at the network level: switching a domain on or off a restricted list, cheap and immediate compared with training, curriculum change or data-privacy vetting. The January block asked nothing of teachers. The May reversal asked considerably more: schools were told to expect resources, shared implementation examples and a classroom discussion toolkit, none of which follows automatically from a network policy alone. A permission and an instructional programme are different kinds of decision, and the chancellor's essay names no budget, training schedule or data-handling standard for the tools students would now be encouraged to use.

What holds and what fails

What holds is the observation that a blanket access rule is a poor proxy for a considered policy in either direction: blocking a domain does not teach the caution it invokes, and unblocking it does not deliver the guidance a chancellor's essay promises. What is likely to fail, editorially, is treating either date as the moment the question was settled; the sources describe a change in official posture, not one verified by classroom data. A different district reading this record should notice that the reversal's authority rests on the chancellor's own account of his reasoning, a valid but singular kind of evidence.

• Did the original restriction target a stated risk, or a general unease that a rule change simply relabels?
• What resourcing, beyond a toolkit, would make the reversal's promised support real in a specific school?
• Whose account of what changed is being trusted, and is it backed by anything beyond that same office's own statement?

The New York City record is unusually candid because a sitting chancellor put his reasoning in writing twice, five months apart. By the 2026-27 school year, the department's own current guidance shows the pendulum swinging again, toward a moratorium on student-facing generative AI in grades 2K-8 and only limited, vetted use in high school, as the page states it stood on 16 September 2026 — itself evidence that neither the 2023 block nor its reversal settled the underlying question.