New York and Los Angeles pull back on classroom AI, without ever measuring whether it worked

New York City and Los Angeles run the two largest public school systems in the United States. In the same first week of September, both pulled back hard on generative artificial intelligence in the classroom, and both cast the decision as caution rather than technophobia. Neither district, in explaining itself, could point to a study showing the technology had done students any good. That absence is the real story.

New York City Public Schools confirmed a joint announcement from Mayor Zohran Mamdani and Schools Chancellor Kamar Samuels that student-facing generative AI is barred outright in grades 2-K through 8 for the school year that begins September 10, a freeze the district says covers roughly two-thirds of its enrollment. High schools keep a narrow opening: five vetted pilots, capped at 50,000 students, covering tools such as Quill for English instruction and Edia for math, each restricted to a set number of minutes per week. The district's own published guidance pairs the ban with new screen-time ceilings — no personal devices before third grade, 30 minutes a day through fifth grade, 45 through eighth — and exempts assistive technology used under students' disability or language-support plans.

A ban written into the firewall, not the policy manual

Los Angeles Unified moved with less ceremony. Rather than a new board policy, the district simply reconfigured its Lightspeed web-filtering system to block any site or app categorized as generative AI on school-issued devices, ending an arrangement under which students 13 and older who had completed a digital-citizenship course could use approved tools. Los Angeles blocked the technology at the network level, a filtering change rather than a deliberated policy, which meant the restriction landed before most people who oversee the district had been told it was coming. At the first meeting of the district's Generative AI Ad Hoc Committee, board member Nick Melvoin remarked on the parents' visible surprise in the room and said the change was not well publicized; committee chair Kelly Gonez confirmed the pause was not meant to be permanent. The episode caught its own school board and parents off guard, even as one parent advocacy group, Schools Beyond Screens, welcomed it as districts "reclaiming" schools from corporate influence.

Both committees — New York's new Technology in Schools Coalition and Los Angeles's ad hoc panel — are charged with producing recommendations by the end of the 2026-27 school year. In other words, neither system is claiming to have already answered the question of whether the technology helps or harms learning. They are pausing in order to go find out, three years after many of the same districts began rolling the tools out.

The collapse that should have been a warning

Los Angeles Unified has been here before. In June 2023 the district signed a five-year, $6 million contract with the startup AllHere Education to build "Ed," a chatbot pitched as a personalized academic and mental-health assistant for students and families. Ed launched in March 2024 to national press attention. By mid-June, AllHere had furloughed most of its staff, LAUSD pulled the chatbot offline for lack of anyone to supervise it, and the district had already paid out roughly half the contract's value. AllHere filed for bankruptcy that September.

Two months later, federal prosecutors in the Southern District of New York charged the company's founder with securities fraud, wire fraud and aggravated identity theft, alleging she fabricated a financial consultant's identity and emails to convince investors the company had far more revenue and far more school-district partnerships, including in New York City and Boston, than it actually did. Prosecutors put the alleged fraud at close to $10 million. The founder was arrested in North Carolina.

That collapse happened inside the same buying cycle these two moratoria now describe: a district under pressure to look forward-leaning on technology, a vendor whose claims were never independently verified before the contract was signed, and a rollout to actual children that outran any capacity to check whether the tool worked or was even still operating. The lesson available in 2024 — verify before you deploy, and build in a way to measure results — is essentially the lesson both districts are only formalizing now, in blunter form, as a blanket pause.

"Innovation does not mean more technology, and over the next year, we will lead with evidence," Chancellor Samuels said in announcing the moratorium.

Washington is still pushing the other way

The retreat in New York and Los Angeles runs directly against federal policy. The U.S. Department of Education finalized a new supplemental grant priority this spring that rewards discretionary-grant applicants for projects expanding AI literacy and "appropriate and ethical use" of the technology in schools, effective in May. The rule sets no comparable requirement that grantees demonstrate the tools improve outcomes before that money flows; it asks districts to promote adoption, not to prove benefit. The country's two largest school systems are now insisting on the opposite standard for themselves, using their own students as the group that will not be experimented on again without evidence in hand.

That gap matters beyond these two cities. Virginia's governor has already urged more districts to follow suit and restrict chatbot access, and dozens of state education agencies are drafting their own AI guidance with no settled federal benchmark to work from. Superintendents in smaller districts, lacking New York's compliance staff or Los Angeles's network-filtering infrastructure, are watching two systems that spent heavily to get ahead of this technology now spend just as much effort walking it back.

What the pause actually admits

The more defensible way to describe what happened this month is not that AI proved harmful in classrooms — the evidence for harm is no more settled than the evidence for benefit — but that two of the country's most closely watched school systems have concluded, after roughly three years of chatbot pilots, that they cannot say whether any of it worked. A moratorium justified by "we don't yet know" is a reasonable, even overdue, response to that admission. It is a much less reassuring one from institutions that spent millions finding out the hard way, in Los Angeles's case, exactly what happens when the checking is skipped. The ed-tech vendors named in New York's surviving pilots — Quill, Edia, Brisk Teaching, Playlab, and Intel's AI-Ready Schools program — are now the test case for whether a narrower, better-monitored rollout can produce the evidence that the broader one never did. If it cannot, the pause due to expire at the end of this school year may simply become the policy.