Three Years to Adopt, One Summer to Undo: Banning Student AI

New York City Mayor Zohran Kwame Mamdani and Schools Chancellor Kamar H. Samuels announced on 2 September 2026 a one-year moratorium on student-facing generative AI for grades 2-K through eighth grade, reaching close to 600,000 of the city's 793,300 K-12 pupils, while Los Angeles Unified confirmed the same day that generative AI is blocked for all students on district-issued devices for 2026-27. New York's policy exempts assistive technology for students with disabilities, multilingual learner tools, required computer-based assessments, career and technical education and centrally approved instructional programmes, and it prohibits companion chatbots outright across all grades with no carve-out. The two largest U.S. school systems reversed three-year-old technology policies in a single Wednesday, with LA board member Nick Melvoin saying of the district's undisclosed filter, "I don't think it's well-publicised.

Three Years to Adopt, One Summer to Undo: Banning Student AI The block is enforced by web-filtering software called Lightspeed. A pupil at a Los Angeles Unified school opens a district Chromebook, types a chatbot's address into the browser, and gets a page saying the category is not permitted. If she picks up her own laptop but is still signed into her district profile, the same page appears. There was no letter home, no press conference. The filter simply went up, and the school year began. It stayed quiet until the morning of 2 September 2026, when the district's new Generative AI Ad Hoc Committee met at 333 South Beaudry Avenue and a room full of parents worked out, in real time, that what they had come to argue about had already happened. “Just based on the facial expressions of folks in this room, many parents, some of whom work at the district, some of whom don't, I don't think it's well-publicised,” board member Nick Melvoin said. “I would encourage us to do more.” Kelly Gonez, who chairs the committee, confirmed the state of play in one sentence. “Currently, there is no access to generative AI for any student.” That same day in New York, Mayor Zohran Kwame Mamdani and Schools Chancellor Kamar H. Samuels announced what the mayor's office called the broadest generative AI moratorium in the nation. The two largest school systems in the United States, between them responsible for well over a million children, spent a single Wednesday reversing a technology policy they had spent three years building. Nobody has explained why it took three years to adopt and one summer to undo. That asymmetry is the story. What Was Actually Blocked and What Quietly Was Not Start with the numbers, because several of the ones in circulation are wrong. New York's moratorium is not a ban on artificial intelligence in schools, and it does not cover a million students. It suspends student-facing generative AI for 2-K through eighth grade for one year, reaching close to 600,000 pupils, roughly two-thirds of enrolment. State education data recorded 793,300 pupils in the city's K-12 grades in 2025-26, with the widely used 900,000 figure arriving once 3-K and pre-K are included. It is comfortably the largest district in the country. It is not a million children, and the moratorium does not reach all of them. The exemptions matter more than the headline. Assistive technology for pupils with disabilities is exempt, as are tools for multilingual learners, required computer-based assessments, career and technical education, and, in a phrase doing enormous load-bearing work, centrally approved instructional programmes. Teachers may keep using generative AI for planning, though not to grade work, determine promotion, or counsel a child in crisis. Officials intend to switch off AI features inside 38 citywide software contracts, including the Amira reading tutor used in 222 schools, Houghton Mifflin Harcourt's Into Reading, and Google Gemini on Chromebooks. High schools are exempted almost entirely. Up to 50,000 pupils may join pilots of five vetted products, with startlingly specific time caps of fifteen minutes a week for Quill and twenty for Edia. Every high schooler will sit two 45-minute AI critical thinking modules a year. Screen time gets its own ladder: no one-to-one devices through second grade, thirty minutes a day in grades three to five, forty-five in middle school, and no guidance at all for high school. One provision reaches every grade with no carve-out, and it received a single sentence in the announcement and almost no coverage. Companion chatbots are prohibited outright, across all grades. Of everything in the policy, that is the piece with the clearest evidentiary case behind it, and it is the piece nobody discussed. Los Angeles is blunter and narrower at once. There is no grade cut-off and no pilot programme: generative AI is blocked for all students on district-issued devices for 2026-27, full stop. Teachers keep access. The previous policy let pupils aged thirteen and over use approved tools at a teacher's direction after completing a digital citizenship lesson. That is now void. The affected population has been reported at roughly 378,000. Higher figures circulate, including 497,105 for 2025-26 in the state's data portal and a count approaching 565,000 that sweeps in independently operated charter schools, early education and adult programmes. The often-quoted 565,000 is not the number of children hitting the block page. Both cities were widely reported as acting on 4 September. Both acted on 2 September, and Los Angeles was arguably first, since its filter had been running since the start of term. Thirteen Undergraduates and a Policy for Four-Year-Olds The evidentiary case for these decisions leans on three recent preprints, all worth reading. None says quite what it is being used to say, and not one involves a child. The first, posted to arXiv on 24 April 2026 by Abeer M. Hasan and Sayed A. Mostafa, is “Perceptions and Utilization of GenAI Tools among Data Science Students and Faculty”. It surveyed 119 students and 14 faculty at a historically Black college or university. Students used generative AI heavily, mostly ChatGPT, for coding and writing, and reported limited confidence in interpreting what it gave back. The authors conclude there is “a gap between AI adoption and AI literacy” and recommend structured training. It is a useful survey. It contains no measure of critical thinking, no control group and no children. It cannot show that overreliance reduces critical thinking, and it does not claim to. Its recommendation is more training, not less access. The second, posted on 29 June 2026 by Keith Tran, Colton Harper and Thomas Price, is titled “Why Put in This Much Effort?” and asks how the availability of AI shapes motivation in introductory programming. The method is thirteen semi-structured interviews with engineering majors in a MATLAB course with a bespoke chatbot. The findings are genuinely interesting: students questioned whether their time was well spent, doubted the lasting utility of the skill, reported less satisfaction when the machine bypassed productive struggle, and described a confidence that evaporated when the tool was unavailable. But it measures nothing. It is a qualitative study of thirteen undergraduates, reporting no effect sizes because it was never designed to. Its concluding sentence points in precisely the opposite direction to the one the policy debate has assigned it. “Our findings,” the authors write, “complicate the assumption that students need external constraints to protect their learning.” The students who managed the tension found motivation in the process of learning itself. The paper's recommendation is to redesign courses to value how students learn rather than what they produce. That is an argument about assessment design, not an argument for a filter. The third is the most interesting and the most badly misread. Shahin Hossain's “Four Types of LLM Reliance and Their Predictors Among Undergraduate Writers”, posted on 27 June 2026, draws on 382 undergraduates at a public minority-serving R1 university, plus 14 interviews and 396 open-ended responses. It identifies four reliance types: Strategic at 34.3 per cent, Instrumental at 30.9, Dialogic at 30.4 and Dependent at just 4.5. A further 13 per cent or so declined to use AI on ethical rather than practical grounds, a category existing frameworks miss entirely. The paper does not find that students who use large language models show lower critical thinking than those who do not. Its striking result is almost the inverse. Strategic users, the ones engaging most deliberately and showing what the author calls “the greatest independent thinking”, scored lowest on standard outcome measures. Hossain's reading is that the instruments are broken. They “index AI's contribution rather than writing quality”, rewarding dependence and penalising students doing their own thinking. That is a paper about measurement failure at university level. It has been recruited into a debate about eight-year-olds. This is not a small problem. Two of the largest school systems in the country have restricted a technology for children as young as four, and the research most frequently invoked studies undergraduates, in samples of 13, 119 and 382. There may be good reasons to keep chatbots away from a six-year-old. These papers are not among them. Cognitive Debt and Its Discontents The wider literature is stronger, though nowhere near as strong as the citations suggest. The paper everyone reaches for is “Your Brain on ChatGPT”, released as a preprint in June 2025 by Nataliya Kosmyna, Pattie Maes and colleagues at the MIT Media Lab. Participants wrote timed SAT-style argumentative essays while wearing EEG caps, split into three conditions: ChatGPT, a search engine, or nothing but their own head. Brain-only writers showed the strongest and most distributed neural connectivity, search engine users sat in the middle, the language model group weakest. Model users struggled to quote their own essays back. The paper supplied the phrase “cognitive debt” that has since escaped into every school board meeting in the country. The caveats are considerable, and they compound. Fifty-four people took part in the first three sessions, divided across the three conditions: roughly eighteen per condition, already below the norm for EEG work in a field worrying about underpowered designs. Then the attrition. Only eighteen of those fifty-four returned for the fourth session, the one in which participants swapped conditions, which leaves about six people per group. The thinnest part of the study is the part carrying the most interpretive weight. Participants were Boston-area adults aged eighteen to thirty-nine writing twenty-minute essays. The paper was a preprint at release, not peer reviewed, and a formal comment challenging its interpretation has since appeared. Whatever “cognitive debt” describes, it was not observed in a child, in a classroom, over a term. Michael Gerlich's 2025 study in Societies, the most cited quantitative claim about cognitive offloading, drew on 666 participants and reported a significant negative correlation between frequent AI use and critical thinking, with younger participants faring worst. It is correlational. It cannot establish that AI use degrades thinking rather than that people who think less reach for AI more, and the journal issued a correction in September 2025 after a table was duplicated. The most methodologically satisfying entry is Yizhou Fan, Dragan Gasevic and colleagues in the British Journal of Educational Technology, whose paper gave the field the phrase “metacognitive laziness”. Their learners using ChatGPT improved short-term task performance without gains in intrinsic motivation, knowledge acquisition or transfer, and showed fewer metacognitive processes such as evaluation and orientation than those working with human experts or a checklist. That is a real, peer-reviewed finding about mechanism, and it points somewhere specific: the problem is not the machine's presence but the collapse of the evaluative step, which is a pedagogy problem rather than a network filter problem. Taken together the evidence supports a cautious, adult-supervised, assessment-redesigned approach with young learners. It does not license the confidence with which either the adoption or the restriction was carried out. The Sun With a Face To understand why Los Angeles reached for the filter, you have to understand what it is still recovering from. On 20 March 2024, at the Edward R. Roybal Learning Center, Superintendent Alberto Carvalho unveiled Ed, an AI chatbot personified as a cheerful round sun, built to pull student records, assignments, grades, recommendations and mental health referrals into a single portal. “Simply put, Ed relies on the information that the district already possesses, analyses, personalises it to the needs of each student and then builds a pathway,” Carvalho told reporters. The contract with a Boston start-up called AllHere Education, signed on 20 June 2023, was worth up to about $6.2 million over five years. The district paid roughly $3 million. Ed was switched off on 14 June 2024, less than three months after launch, when AllHere furloughed most of its staff. Chief of Special Education Tony Aguilar said the district had no warning until the day it happened. AllHere filed for Chapter 7 bankruptcy later that year. On 19 November 2024, federal agents arrested the company's founder and chief executive, Joanna Smith-Griffin; prosecutors in the Southern District of New York charged her with securities fraud, wire fraud and aggravated identity theft, alleging she had told investors in 2021 that AllHere had generated about $3.7 million in revenue, including contracts with the New York City and Atlanta districts, when the true figure was around $11,000 and the contracts did not exist. Those remain allegations. The consequences travelled upwards, though not by any route that has yet been established in public. In February 2026 federal agents arrived at Carvalho's home and district office in the early morning, and at the Miami property of a Florida consultant with long-standing ties to him who had helped bring the technology to the district. Reporting at the time indicated the raids may be linked to the AllHere deal; that link has been suggested rather than demonstrated. A source with direct knowledge has since said the investigation concerns allegations of kickbacks dating to Carvalho's earlier tenure running the Miami school district, and that it predates the current administration. He is not named in the case against Smith-Griffin and has not been charged with any crime. He was placed on paid leave on 27 February and resigned on 21 June 2026, saying that “placing students first has always guided my work”. Andrés Chait succeeded him on an acting basis. None of which requires Carvalho to have done anything wrong. The argument does not turn on what the investigation concludes, because what follows is a sequence of institutional events rather than a verdict: superintendent champions a chatbot, contract signed, product launched with a press event, dead in eleven weeks, vendor bankrupt, founder indicted, superintendent raided and gone, and two months after his resignation the district blocks the technology for every child it educates. That is not a district responding to research. It is a district responding to a scandal. Everything Old Is Procured Again Los Angeles had run this loop before, in a version so similar that the main difference is the price. In 2013, Superintendent John Deasy launched a plan to put an iPad loaded with Pearson curriculum into the hands of every student, some 650,000 devices, at a projected cost approaching $1.3 billion. The rollout collapsed almost immediately. Pupils bypassed the security restrictions within days, teachers had not been trained, and the content was incomplete. A federal review later noted that Deasy had given the district only a few months to produce a plan before putting the work out to bid, and emails surfaced showing he and a deputy had been in close contact with Apple and Pearson before the contract was awarded. The FBI seized twenty boxes of records in December 2014. Deasy had resigned that October, and his successor Ramon Cortines cancelled the Pearson contract the day after the seizure. Federal investigators closed the inquiry without charges in February 2017. Set the two episodes side by side and the pattern is legible. A charismatic superintendent identifies a transformative technology. Procurement moves faster than evaluation. The product ships to hundreds of thousands of children with no pilot worth the name. It fails within a school year. Federal law enforcement arrives. The leader departs. The district retreats. Twice in thirteen years, in the same building, the reversal has been triggered by a criminal investigation rather than by a study. Rachel White, associate professor of educational leadership at the University of Texas at Austin, put the dynamic carefully when the district's leadership imploded in June. Carvalho “was always willing to make big, ambitious bets”, she said, “and that boldness can sometimes cut both ways”. Natalie Milman of George Washington University offered the mundane fix that neither the iPad programme nor Ed received: “Before rolling out any tool, educators should conduct some testing with a smaller group, getting some feedback from the actual people who will be using it, and then getting that public comment.” Ed went from contract to citywide launch without that. So did the moratorium that replaced it. The Chancellor Who Changed His Mind New York's record is a purer specimen, because it contains a completed round trip. In January 2023, weeks after ChatGPT became generally available, the New York City education department added it to the blocked list on school networks and devices, citing academic integrity and accuracy. The ban lasted four months. On 18 May 2023, Chancellor David Banks announced the reversal in a Chalkbeat op-ed, the same institution discussing the same technology from the opposite direction. “The knee-jerk fear and risk overlooked the potential of generative AI to support students and teachers,” Banks wrote, “as well as the reality that our students are participating in and will work in a world where understanding generative AI is crucial.” He promised the department would “encourage and support our educators and students as they learn about and explore this game-changing technology”. Three years and three months later, a different mayor and chancellor announced the country's broadest moratorium. “Children need teachers and human connection in order to learn and grow,” Mamdani said. Samuels framed it as parental protection: “As Chancellor and as a parent, I believe our job is to protect what makes learning work.” Both statements are perfectly reasonable. So was Banks's. That is the problem. In under four years the same school system has banned generative AI, unbanned it while calling the ban a knee-jerk reaction, spent three years encouraging it, and banned it again for two-thirds of its pupils. At no point did a decisive piece of evidence about children and chatbots arrive. What arrived, in order, was a moral panic, a wave of vendor and federal enthusiasm, and then a political constituency of organised parents. The enthusiasm phase is worth pricing. In April 2025 an executive order titled “Advancing Artificial Intelligence Education for American Youth” created a White House Task Force on AI Education, directing agencies to build partnerships teaching K-12 pupils AI literacy. In July 2025 the American Federation of Teachers, with the United Federation of Teachers, announced a National Academy for AI Instruction underwritten by $23 million over five years from Microsoft, OpenAI and Anthropic, aiming to train 400,000 educators from a Manhattan facility, beginning with New York City. The Academy is a few kilometres from the chancellor who has now suspended student-facing generative AI for 600,000 children. The union whose president has publicly criticised the moratorium's enforcement gaps is the same union that co-founded it. Seven days after the moratorium, that observation acquired a sharper edge. On 9 September 2026, Randi Weingarten of the AFT, Michael Mulgrew of the UFT and Brad Smith, vice chair of Microsoft, jointly announced a National AI Safety and Privacy Standard for schools, framed as legally enforceable rather than advisory. Student and educator data cannot be used to train AI models, sold, or repurposed. Schools keep control over how long data is retained and when it is deleted. AI systems must carry human oversight and cannot make decisions without it. Families must be given clear information about how AI is being used. “Anything less than legally enforceable provisions is simply a wish list,” Weingarten said. Mulgrew presented it as a debt being settled: “This agreement gives families and school districts the information and privacy protections they have been demanding.” Microsoft's own announcement says the standard builds on New York City's screens and AI policy. The Academy and the Standard are separate initiatives, and that distinction is worth stating plainly. The sequence is still striking. Within a week of telling the city it had not explained how it would make purchased products comply, the union president was co-announcing enforceable national protections with one of the companies underwriting the Academy. The terms on which several hundred thousand children meet generative AI are being set in the language of data retention and model training, because that is the language a negotiation between a union and a vendor produces. None of it is a finding about what the technology does to a nine-year-old. The vacuum is being filled by the parties who were already in the room. Meanwhile the adoption curve ran fast and unevenly. RAND's surveys found the share of districts training teachers on AI more than doubled between 2023-24 and 2024-25, from 23 per cent to 48 per cent, while only about 35 per cent of district leaders reported providing pupils with any AI training. Nearly 60 per cent of principals reported using AI in their own work; only 18 per cent said their district had issued guidance on staff or student use. The technology entered American schools through the staffroom, without a policy, and the policy is arriving three years late in the form of a switch being flipped. The Strongest Case Against the Filter The argument against these moratoriums is not vendor lobbying and should not be dismissed as such. It has three legs. The first is equity, and it is the one that ought to worry the districts most. Blocking generative AI on district devices does not remove it from children's lives; it removes it from the children whose only reliable device is the district's. Common Sense Media's 2026 census of 1,204 American children aged nine to seventeen found 86 per cent already using AI in some form, close to a quarter of them daily, and more than four in ten saying no parent had ever talked to them about AI safety. Pew Research Center has consistently found that teenagers in lower-income households are less likely than wealthier peers to have a home computer or tablet, even as smartphone ownership approaches universal. The distributional consequence is straightforward. A child in Brentwood or Brooklyn Heights with a family laptop and a parent who has views about prompt engineering will keep using these tools at home, unsupervised, with no critical thinking module attached. A child whose access runs through a district Chromebook will meet the technology first at sixteen, or at work. RAND's figures already show the gradient: by the start of 2025-26, district leaders estimated that almost all low-poverty districts would have trained teachers on AI, against about six in ten high-poverty districts. A blanket device-level block does not close that gap. It formalises it. Richard Buery Jr, chief executive of the anti-poverty organisation Robin Hood, made the point precisely. The moratorium for younger grades, he said, is “not the right call”, and should not extend beyond the coming school year. “In an understandable effort to protect students, I fear we're rejecting uses that could help them.” The second leg is that blocking a tool is not the same as teaching judgement about it, and only one of those is education. New York gestures at this with two annual 45-minute modules for high schoolers, ninety minutes a year of formal instruction about the most consequential general-purpose technology of their working lives. There is no equivalent for K-8 at all. BetaNYC's Noel Hidalgo noted that the policy provides no AI literacy instruction for the banned grades. Hossain's finding is directly relevant: what predicted the type of reliance a student adopted was not access, but AI literacy. Value and cost beliefs predicted how much they used it; literacy predicted whether they used it well. A policy that reduces access without increasing literacy is optimising the wrong variable. The third leg is that the ban may not be a ban. The Parents' Coalition for Student Privacy argues that most real classroom applications fall into a vague middle tier left to teacher discretion, and organiser Kelly Clancy says the policy “may be less restrictive than it appears”. Liat Olenick of the AI Moratorium Coalition, which wanted two years rather than one, was blunter: “There are a lot of holes in the policy. If we're honest, there's a lot of unanswered questions. Too many exceptions.” UFT president Michael Mulgrew said the department has not explained how it will make purchased products comply, and that educators should not have to “figure out after the fact” whether their school is in breach. There is no published complaint mechanism, no public directory of which tools run in which school, and vendor non-disclosure agreements shield product details from scrutiny. State Comptroller Thomas DiNapoli's August audit found the city holds no complete inventory of its AI systems. In Los Angeles, Anya Meksin, deputy director of the campaign group Schools Beyond Screens and a district parent, sits on the committee reviewing the moratorium and does not think a pause is the point. “I think the current policy needs to be discarded completely,” she said, “and we need to start from scratch with the foundational question of what is the benefit of these tools in education?” That question was never answered before the tools arrived. It is not being answered now that they have gone. The Asymmetry Nobody Is Measuring What the two speeds reveal is not what either side wants it to be. Adoption was fast because adoption is a purchase. It needs a superintendent, a procurement officer, a vendor with a demo and a board vote most people in the room do not understand. Ed went from contract signature to a launch event for hundreds of thousands of children in nine months. Nobody had to be persuaded of anything difficult, because the persuading happened in a few meetings with a few people. Restriction was slow because restriction is a politics. It requires a constituency. In Los Angeles that was organised parents who spent the spring campaigning on screens and were rewarded in June with a board vote phasing device use by grade. In New York it was a coalition of parents, a new mayor and a new chancellor with an interest in demonstrating they were not the last administration. The evidence base moved barely at all. What moved was the coalition. Chicago is that mechanism happening in real time and on a ballot. Of the 42 candidates standing for the city's school board this autumn, 24 back a temporary moratorium on AI as part of wider limits on screen time, including all fifteen endorsed by the local teachers' union. Restriction has stopped being a position a district arrives at and become a position a candidate runs on. No new study produced that. A constituency did. That is why the reversal looks abrupt and is not. The chatbot has been in American classrooms for three and a half years, and the restriction took that long to assemble because it had to be built out of people rather than contracts. People organise more slowly than budgets spend. And because the constituency is local, so is the direction. While New York and Los Angeles were switching the technology off, Miami-Dade County Public Schools, the third-largest district in the country, signed a partnership with Google to bring Gemini into its classrooms. The first, second and third largest school systems in the United States looked at the same absent evidence base in the same few weeks and moved decisively in opposite directions. If research were setting the direction, that could not happen. Districts of that size do not read different literatures. They have different politics, different scandals and different vendors in the room. Which leaves the genuinely damning observation. Both decisions, the adoption and the reversal, were made without adequate evidence about what generative AI does to a nine-year-old, because that evidence still barely exists. The preprints being waved around study undergraduates. The EEG study has eighteen adults per condition, and six by its final session. Nobody has run the long-run trial with children that would justify either the enthusiasm of 2023 or the certainty of 2026, and both institutions acted at full scale anyway, in opposite directions, on the same absent evidence. The problem is not that districts were too fast then and too slow now. It is that both speeds were set by something other than knowing. Building an Institution That Can Change Its Mind Cheaply If there is a defensible version of what these two cities have just done, it lies in a distinction neither has made explicitly: the difference between a decision that is reversible and one that is not. New York's moratorium runs for one year and comes with a Technology in Schools Coalition of students, educators, parents, officials, union partners and researchers, charged with assessing impact and publishing recommendations. Los Angeles has a committee, six scheduled meetings running from September 2026 to April 2027, and an obligation to report to the board before the year is out. The dates are 2 September, 21 October and 9 December, then 20 January, 3 March and 7 April. Which means the meeting this article opens on was the first of the six: the parents working out in real time that the block had already gone up were sitting in session one of a process meant to produce, in Gonez's framing, AI policy recommendations the full board could vote on next year. Gonez has been careful not to oversell the permanence. “I wouldn't expect necessarily that there would be a pause for forever,” she said. “But this provides us the opportunity to really think through what does our policy look like moving forward.” Time-limited, reviewable, with a named body and a reporting date is a better decision architecture than a five-year contract signed after a demo. It is a shame it took two federal investigations to arrive at it. What would make it credible is a handful of unglamorous commitments. Publish the inventory of which AI-containing products run in which schools, which is exactly what DiNapoli found New York lacks. Pilot before scale, with the small-group testing and public comment Milman described, rather than a launch event at a named high school. Write reversibility into contracts, so a vendor's failure does not become an emergency for 378,000 children. Fund the literacy alongside the restriction, because Hossain's data says literacy determines whether reliance becomes dependence. And commission the research on the actual population, because a district of half a million pupils is better placed to run that study than the universities currently supplying the citations. Melvoin's complaint at that first meeting was, in the end, the most important thing said in either city. The district changed the terms on which several hundred thousand children encounter the defining technology of their era, and did not tell their parents. The substance may well turn out to be right. The process was the one that bought the iPads and the sun with a face: a decision taken by a few people, at speed, and explained afterwards. A ten-year-old in Los Angeles will hit a Lightspeed block page this week. In four years she will be expected to use these systems competently in a workplace that assumes she grew up with them. Between those two facts sits an institution that has now shown, twice in three years in one city and twice in thirteen in the other, that it can move extremely fast in whichever direction the last scandal points. Speed is not the failure. Direction without evidence is. The block page, whatever else it does, buys time to find some. Sources and References 1. NYC Mayor's Office, “Mayor Mamdani and Chancellor Samuels Put Students First with Nation's Broadest Generative AI Moratorium in Schools”, 2 September 2026. https://www.nyc.gov/mayors-office/news/2026/09/mayor-mamdani-and-chancellor-samuels-put-students-first-with-nat https://www.nyc.gov/mayors-office/news/2026/09/mayor-mamdani-and-chancellor-samuels-put-students-first-with-nat 2. Chalkbeat New York, “NYC's new school AI ban explained: What students and teachers need to know”, 2 September 2026. https://www.chalkbeat.org/newyork/2026/09/02/what-to-know-about-nyc-public-schools-generative-ai-ban-screen-time-limits/ https://www.chalkbeat.org/newyork/2026/09/02/what-to-know-about-nyc-public-schools-generative-ai-ban-screen-time-limits/ 3. Noel Hidalgo, “NYC's DOE AI Ban? Here Is What the Policy Actually Says”, BetaNYC, 4 September 2026. https://www.beta.nyc/2026/09/04/nyc-paused-ai-in-its-classrooms-here-is-what-the-policy-actually-says/ https://www.beta.nyc/2026/09/04/nyc-paused-ai-in-its-classrooms-here-is-what-the-policy-actually-says/ 4. Governing, “New York City Put Limits on AI in Schools. Are They Enough?”, September 2026. https://www.governing.com/artificial-intelligence/new-york-city-put-limits-on-ai-in-schools-are-they-enough https://www.governing.com/artificial-intelligence/new-york-city-put-limits-on-ai-in-schools-are-they-enough 5. LAist, “How will LAUSD students use AI? Here's how you can help decide”, September 2026. https://laist.com/news/education/lausd-ai-committee-first-meeting-tech-in-schools https://laist.com/news/education/lausd-ai-committee-first-meeting-tech-in-schools 6. Education Week, “LAUSD Adopts Moratorium on Student AI Use, Another Big Move to Limit Tech Use”, September 2026. https://www.edweek.org/technology/lausd-adopts-moratorium-on-student-ai-use-another-big-move-to-limit-tech-use/2026/09 https://www.edweek.org/technology/lausd-adopts-moratorium-on-student-ai-use-another-big-move-to-limit-tech-use/2026/09 7. California Department of Education, Enrollment by Grade, Los Angeles Unified, 2025-26. https://dq.cde.ca.gov/dataquest/dqcensus/enrgrdlevels.aspx?agglevel=District&year=2025-26&cds=1964733 https://dq.cde.ca.gov/dataquest/dqcensus/enrgrdlevels.aspx?agglevel=District&year=2025-26&cds=1964733 8. New York State Education Department, NYC Public Schools enrolment data, 2025. https://data.nysed.gov/enrollment.php?instid=7889678368 https://data.nysed.gov/enrollment.php?instid=7889678368 9. Abeer M. Hasan and Sayed A. Mostafa, “Perceptions and Utilization of GenAI Tools among Data Science Students and Faculty”, arXiv:2604.22998, 24 April 2026. https://arxiv.org/abs/2604.22998 https://arxiv.org/abs/2604.22998 10. Keith Tran, Colton Harper and Thomas Price, “Why Put in This Much Effort?”, arXiv:2606.30480, 29 June 2026. https://arxiv.org/abs/2606.30480 https://arxiv.org/abs/2606.30480 11. Shahin Hossain, “Four Types of LLM Reliance and Their Predictors Among Undergraduate Writers”, arXiv:2606.28749, 27 June 2026. https://arxiv.org/abs/2606.28749 https://arxiv.org/abs/2606.28749 12. Nataliya Kosmyna, Pattie Maes and colleagues, “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task”, MIT Media Lab, arXiv:2506.08872, 10 June 2025. https://www.media.mit.edu/publications/your-brain-on-chatgpt/ https://www.media.mit.edu/publications/your-brain-on-chatgpt/ 13. Michael Gerlich, “AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking”, Societies, vol. 15, no. 1, art. 6, 2025, with Correction, 10 September 2025. https://www.mdpi.com/2075-4698/15/1/6 https://www.mdpi.com/2075-4698/15/1/6 14. Yizhou Fan, Dragan Gasevic and colleagues, “Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance”, British Journal of Educational Technology, vol. 56, no. 2, 2025, pp. 489-530. https://bera-journals.onlinelibrary.wiley.com/doi/10.1111/bjet.13544 https://bera-journals.onlinelibrary.wiley.com/doi/10.1111/bjet.13544 15. Government Technology, “LAUSD Shutters New AI Chatbot As Developer Goes Out of Business”, 2024. https://www.govtech.com/education/k-12/lausd-shutters-new-ai-chatbot-when-developer-goes-out-of-business https://www.govtech.com/education/k-12/lausd-shutters-new-ai-chatbot-when-developer-goes-out-of-business 16. US Attorney's Office, Southern District of New York, “CEO Of Artificial Intelligence Startup Company Charged With Defrauding Investors”, 19 November 2024. https://www.justice.gov/usao-sdny/pr/ceo-artificial-intelligence-startup-company-charged-defrauding-investors https://www.justice.gov/usao-sdny/pr/ceo-artificial-intelligence-startup-company-charged-defrauding-investors 17. EdSource, “FBI raid of Los Angeles Superintendent Carvalho's home, office may be linked to student AI chatbot”, February 2026. https://edsource.org/2026/fbi-raids-home-and-office-of-los-angeles-unified-superintendent-carvalho/752038 https://edsource.org/2026/fbi-raids-home-and-office-of-los-angeles-unified-superintendent-carvalho/752038 18. Education Week, “Lessons Learned About Bold Tech Initiatives From the LAUSD Chief's Departure”, June 2026. https://www.edweek.org/leadership/lessons-learned-about-bold-tech-initiatives-from-the-lausd-chiefs-departure/2026/06 https://www.edweek.org/leadership/lessons-learned-about-bold-tech-initiatives-from-the-lausd-chiefs-departure/2026/06 19. Governing, “A Cautionary Tale for Any Government IT Project: L.A.'s Failed iPad Program”, 2015, https://www.governing.com/archive/gov-tablets-los-angeles-ipad-apple-schools.html https://www.governing.com/archive/gov-tablets-los-angeles-ipad-apple-schools.html ; and EdWeek Market Brief, “Feds Drop Investigation Into Los Angeles District Over $1 Billion iPad Purchase”, February 2017, https://marketbrief.edweek.org/regulation-policy/feds-drop-investigation-into-los-angeles-district-over-1-billion-ipad-purchase/2017/02 https://marketbrief.edweek.org/regulation-policy/feds-drop-investigation-into-los-angeles-district-over-1-billion-ipad-purchase/2017/02 20. David C. Banks, “ChatGPT caught NYC schools off guard. Now, we're determined to embrace its potential”, Chalkbeat New York, 18 May 2023. https://www.chalkbeat.org/newyork/2023/5/18/23727942/chatgpt-nyc-schools-david-banks/ https://www.chalkbeat.org/newyork/2023/5/18/23727942/chatgpt-nyc-schools-david-banks/ 21. American Federation of Teachers, “AFT to Launch National Academy for AI Instruction with Microsoft, OpenAI, Anthropic and United Federation of Teachers”, 8 July 2025. https://www.aft.org/press-release/aft-launch-national-academy-ai-instruction-microsoft-openai-anthropic-and-united https://www.aft.org/press-release/aft-launch-national-academy-ai-instruction-microsoft-openai-anthropic-and-united 22. Microsoft Source, “AFT, UFT and Microsoft announce 'National AI Safety & Privacy Standard' for schools to protect students, families and educators”, 9 September 2026. https://news.microsoft.com/source/2026/09/09/aft-uft-and-microsoft-announce-national-ai-safety-privacy-standard-for-schools-to-protect-students-families-and-educators/ https://news.microsoft.com/source/2026/09/09/aft-uft-and-microsoft-announce-national-ai-safety-privacy-standard-for-schools-to-protect-students-families-and-educators/ 23. RAND Corporation, “AI Use in Schools Is Quickly Increasing but Guidance Lags Behind: Findings from the RAND Survey Panels”, 2025. https://www.rand.org/pubs/research reports/RRA4180-1.html https://www.rand.org/pubs/research reports/RRA4180-1.html 24. Common Sense Media, “The Common Sense Media Census: AI Use by Tweens and Teens, 2026”, 8 June 2026. https://www.commonsensemedia.org/sites/default/files/research/report/2026-ai-use-by-tweens-and-teens-1.pdf https://www.commonsensemedia.org/sites/default/files/research/report/2026-ai-use-by-tweens-and-teens-1.pdf 25. Ray Schroeder, “Big School Districts Pause AI Access at School”, Inside Higher Ed, 16 September 2026. https://www.insidehighered.com/opinion/columns/online-trending-now/2026/09/16/big-school-districts-pause-ai-access-school https://www.insidehighered.com/opinion/columns/online-trending-now/2026/09/16/big-school-districts-pause-ai-access-school Tim Green UK-based Systems Theorist & Independent Technology Writer Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk https://smarterarticles.co.uk , challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship. His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities. ORCID: 0009-0002-0156-9795 https://orcid.org/0009-0002-0156-9795 Email: tim@smarterarticles.co.uk mailto:tim@smarterarticles.co.uk Listen to the free weekly SmarterArticles Podcast https://www.smarterarticles.fm