New York City and Los Angeles Restrict AI for Nearly a Million Students in One Week

Framing

For three years this brief has tracked a single direction of travel: governance instruments arriving to authorize and condition AI use in schools. Statutes set floors. State agencies wrote model policies. Boards adopted acceptable use rules. Labor claimed a procurement voice. Students drafted a model act. Utah proved the contracts underneath all of it do not describe what the products actually do. Fulton County wrote the human decision rule into board policy on its own. Every one of those instruments answered the same question: how AI should be used. In a single week, the two largest school systems in the United States answered a different question.

On September 2, 2026, New York City Public Schools removed student-facing generative AI from grades 2-K through 8 for one school year, affecting nearly 600,000 children. The same day, at the kickoff meeting of its Generative AI Ad Hoc Committee, Los Angeles Unified confirmed that its web-filtering software already blocks AI tools for all students at all grade levels on district-issued devices, another 378,000. Together, that is close to a million students, roughly one in fifty public school students in the country, whose access changed inside twenty-four hours. Add the District of Columbia staff-use prohibitions issued September 1 and the Kentucky guidance update issued September 3, and the week produced four instruments from four unrelated bodies pointing in the same direction.

Read the mechanisms, because they are not the same and the difference is the lesson. New York City published a structured policy: named pilots with named time ceilings, a stated list of exceptions, a coalition charged with evaluation, and a commitment to bar any technology tool whose functionality is not mission-critical to learning. That last clause is a procurement standard written as a default of exclusion, and it inverts the burden that districts have carried since 2023. Los Angeles produced the same practical result by setting it in the filter. There was no board vote. The committee chair, a sitting board member, said the restriction was a surprise to her. One district wrote a governance instrument. The other changed a filter and told its board afterward. Both outcomes are defensible. Only one of them is reviewable, appealable, or reversible on a record.

The national data released the same day explains why both districts moved. IBM and Morning Consult surveyed 1,019 K-12 educators and 1,029 parents in July 2026 and found that 76 percent of middle school educators and 73 percent of high school educators reported using AI in the classroom weekly, compared with 45 percent of elementary school educators, while only 20 percent of educators reported extensive AI training. Seventy-seven percent of parents want input into how AI is used in their child's classroom, and 20 percent say they clearly understand the guidance their child already receives. That is the actual condition districts are governing: routine use, thin preparation, and a parent population that has not been told what the rules are. Kentucky named the sharpest edge of it on September 3, warning that products students and adults already use daily will add AI capability on their own schedule, which means an approved-tool list governs a snapshot rather than a system.

The research counterpoint is that all of this is being written on an evidence base thin enough to describe in one sentence. The most relevant peer-reviewed synthesis available is a scoping review of 22 empirical studies. The most relevant elementary trial is a quasi-experiment with 22 children over five school days, in which the front matter and the body text disagree about the direction of the effect. Nothing published inside this brief's fourteen-day research window met the selection standard. No study anywhere measures what a grade-band restriction does to a student outcome. New York City is right that the evidence is thin. It is thin in both directions, which makes restraint a defensible bet rather than a demonstrated correction.

The timing argument is immediate. Board policy calendars open in September. Kansas takes up a draft statewide instructional technology framework on September 8 and 9. Florida votes on its AI rule amendment September 16. Between now and Thanksgiving, your board will receive the New York City policy from a parent, the Los Angeles approach from a technology director, and the IBM training number from a union representative. The cost of waiting is not that you adopt the wrong rule. It is that you adopt someone else's rule without the record that would let you defend or revise it.

Top Research and Policy Signals

1. New York City Removed Student-Facing AI From 600,000 Children and Attached a Mission-Critical Test to Every Remaining Tool

Source type. Municipal executive and school district policy action, announced by press release. Not research, not legislation.

Office of the Mayor of the City of New York. (2026, September 2). Mayor Mamdani and Chancellor Samuels put students first with nation's broadest generative AI moratorium in schools [Press release]. nyc.gov

Mayor Zohran Kwame Mamdani and Schools Chancellor Kamar H. Samuels announced a one-year moratorium on student-facing generative AI, effective in the 2026-2027 school year, covering grades 2-K through 8. The city states that the policy reaches nearly 600,000 students, or two-thirds of the system's enrollment, and covers all software that uses student-facing generative AI. Companion chatbots are prohibited across all grades. Five limited pilots will run for a maximum of 50,000 high school students in general education classes, approximately five percent of the student body, at a maximum of five classes per high school, under direct supervision of a trained educator: Quill for English language arts at no more than 15 minutes per week; Edia for mathematics at no more than 20 minutes per week; Brisk Teaching at 10 to 20 minutes once or twice per week; Playlab at no more than two assignments per marking period; and Intel AI-Ready Schools at one period per week. All high school students receive two 45-minute AI literacy modules per year. Students in grade 2 and below face a daily cap of 30 minutes of one-to-one screen time, with recommended caps of 30 minutes daily in grades 3 through 5 and 45 minutes in grades 6 through 8. Teachers may continue using AI for instructional planning and operational tasks. Exceptions apply for assistive technology used by students with disabilities, multilingual learners, and career readiness programs including computer science. A Technology in Schools Coalition will assess the moratorium and publish recommendations. Separately, all technology tools in the system will undergo review and be barred if functionality is determined not to be mission-critical to learning.

Leadership implication. Take the mission-critical review clause to your cabinet before you take the moratorium to your board. Ask your chief technology officer to produce the current list of every tool in your environment with student-facing generative AI, including features switched on inside platforms you already license, and ask your chief academic officer which of those tools would survive a written test of instructional necessity. Most districts cannot produce the first list in a week, which is the finding. Put the grade-band question on your policy committee agenda this month with a one-page written rationale attached, so that when a parent brings the New York City policy to public comment you are amending a position rather than inventing one at the podium.

2. Los Angeles Blocked AI for 378,000 Students Through a Filter Setting, and Its Own Board Found Out at a Public Meeting

Source type. School district administrative practice, disclosed at a public board committee meeting and reported by named trade and public media. Not a board policy, not a published directive, not research.

Merod, A. (2026, September 3). LAUSD restricts all students from using AI tools. K-12 Dive. k12dive.com  ·  Companion report: Dale, M. (2026, September 2, updated September 3). LAUSD students barred from using AI, to the surprise of the school board and parents. LAist. laist.com

At the September 2, 2026 kickoff meeting of the Los Angeles Unified Generative AI Ad Hoc Committee, district officials confirmed that LAUSD web-filtering software now blocks websites and applications categorized as AI tools on district-issued Chromebooks, with a comparable control for tablets. The restriction covers all students at all grade levels, an estimated 378,000 students, according to the district's 2026-27 Fingertip Facts. It replaces a prior posture under district policy Bulletin 63993 that permitted students aged 13 and older who had completed a digital citizenship lesson to use approved AI tools. The control applies to district-issued devices; officials described no mechanism to prevent AI use on a personal device unless the student is signed in to a district profile. Chief Academic Officer Pia Sadaqatmal is quoted on the record; interim Chief Information Officer Douglas Le and senior director of information technology David Cooper described the filtering mechanism. The Ad Hoc Committee was created by a June 2026 board resolution, the same resolution that established student screen-time limits, and is charged with developing policy recommendations for full board approval by the end of the 2026-27 school year. Committee chair Kelly Gonez, a sitting board member, said the restriction was a surprise to her, though she substantially agreed with it. Board members Nick Melvoin and Tanya Ortiz Franklin also sit on the committee. The next committee meeting is October 21, 2026.

Leadership implication. This is the item to take to your own board, and not because of AI. A control affecting every student in the country's second-largest district was implemented at the level of a filtering category, and elected officials learned of it publicly. Ask your cabinet one question this week: which technology decisions in this district can be made by a system administrator without cabinet review, which require cabinet approval, and which require a board vote? Write the answer down and take it to your board as an information item. If your district has a content-filtering console, someone already holds the power Los Angeles just used. The governance question is not whether they should. It is whether anyone above them knows they can.

3. Seventy-Six Percent of Middle School Educators Report Weekly Classroom AI Use and Twenty Percent Report Extensive Training

Source type. Commissioned national institutional survey. Not peer-reviewed. Released alongside a corporate program launch.

IBM. (2026, September 2). New IBM study finds AI adoption is outpacing K-12 readiness [Press release]. IBM Newsroom. newsroom.ibm.com  ·  Full report: IBM. (2026). AI readiness in U.S. schools. newsroom.ibm.com

Morning Consult, commissioned by IBM, surveyed 2,048 United States adults online in July 2026, comprising 1,019 K-12 education professionals and 1,029 parents of K-12 students, with a reported margin of error of plus or minus 3 percentage points for each sample. Among classroom educators, 76 percent of middle school and 73 percent of high school educators report using AI in their classrooms at least weekly, compared with 45 percent of elementary school educators. Separately, 45 percent of high school educators report AI use daily or almost daily. Only 20 percent of K-12 educators say they have received extensive AI training. The top barriers educators cite to supporting AI literacy are lack of training or professional development at 42 percent and limited AI curriculum or instructional materials at 34 percent. On the parent side, 77 percent say they want input on how AI is used in their child's classroom while only 20 percent clearly understand the AI guidance their child receives at school; 51 percent want clearer guidelines, and 48 percent want greater transparency about classroom AI use. Only 24 percent of educators and 16 percent of parents say the United States education system is adapting very well to AI advances. Classroom teachers name student dependency on AI at 52 percent and cheating or plagiarism at 47 percent as their leading concerns. Parents are more than twice as likely as educators to say AI does not belong in K-12 at this time, with 28 percent saying so versus 12 percent.

Leadership implication. Two numbers belong in your next cabinet meeting and your next board presentation, side by side. Weekly classroom AI use is already the majority condition in middle and high school. Extensive training is a one-in-five condition. That gap is not a communications problem; it is a budget line, and it is the strongest available argument for role-based professional learning rather than a single all-staff session. The parent numbers are the second argument. Seventy-seven percent want input, and 20 percent understand the guidance you have already published, which means your existing communication has failed even where your policy has not. Direct your communications lead to produce a one-page, plain-language parent explanation of what AI your students may use, in what class, and what happens if they misuse it, and put it in the fall newsletter rather than the policy manual.

4. The District of Columbia Put AI Surveillance, Discipline, Teacher Evaluation, and IEP Eligibility in the Same Prohibited Category

Source type. State education agency model policy and press release. Guidance only, not binding, not research.

Office of the State Superintendent of Education. (2026, September 1). OSSE releases AI model policy to guide responsible staff use in schools [Press release]. Government of the District of Columbia. osse.dc.gov

OSSE released its first AI Model Policy for Staff Use, built on a three-tier stoplight framework and offered as guidance rather than legal advice, for local education agencies to customize and adopt as they see fit. The red category prohibits AI use for high-stakes decisions that demand human judgment, specifically naming physical surveillance of students and staff, student discipline decisions, evaluations of teacher performance, and determinations of eligibility for individualized education programs or Section 504 accommodations. The yellow category permits limited use with safeguards, naming monitoring digital activity on agency-issued devices, drafting IEP language, reviewing and grading student work, and supplemental coaching for educators. The green category permits use with awareness and a human in the loop, naming lesson planning, customizing student-facing materials, developing tutoring plans, analyzing data sets, school communications, and logistical operations. OSSE recommends staff use only approved enterprise tools, and names FERPA, COPPA, CIPA, IDEA, HIPAA, and the District Protecting Students Digital Privacy Act of 2016 as the compliance floor for any activity involving personally identifiable information. It recommends staff complete training before using AI, demonstrate AI literacy proficiency, renew training annually, and maintain human accountability for all outputs. The policy responds to a February 2026 OSSE survey of local education agency leaders, which found that only 45 percent had established a staff AI policy. The policy focuses on staff use and does not guide student use or tool procurement.

Leadership implication. The distinction worth stealing is the yellow tier. Most district AI policies are binary, permitted or prohibited, which forces every ambiguous use into the permitted column by default. Direct your policy committee to rewrite your staff AI language in three tiers and to place drafting IEP language and grading student work in the middle tier with a named human decision-maker attached, rather than leaving them unaddressed. Then check whether your own red line covers teacher performance evaluation. If your evaluation instrument or its vendor uses any automated scoring or summarization, that is a bargaining-unit question your human resources director needs to see before your next contract cycle, not after.

5. Kentucky Told Districts the Products They Already Bought Will Become AI Products Without Asking

Source type. State education agency guidance, updated September 3, 2026. Informational guidance from the Office of Education Technology, not a regulation and not binding on districts as drafted. Not research.

Kentucky Department of Education, Office of Education Technology. (2026, September 3). AI in Kentucky K-12, September 2026. education.ky.gov

The Kentucky Department of Education updated its statewide K-12 artificial intelligence guidance page on September 3, 2026. Two provisions carry directly into district practice. On embedded capability, the guidance states that products all students and adults use daily will utilize AI to enhance their efficiency and effectiveness, giving intelligent summaries and meeting notes as examples. On data handling, it instructs users never to share personal or confidential information when using publicly available AI tools such as chatbots and virtual assistants. Specifically, it names the situation of well-intentioned K-12 researchers putting sensitive data into such systems. The page is published by the Office of Education Technology, Division of School Technology Planning and Project Management. It sits alongside the 2024 to 2030 KETS Master Plan, which is a regulation by reference. The AI page itself is guidance, not regulation, and does not carry a compliance deadline.

Leadership implication. Your approved-tool list governs a snapshot. Kentucky just said so in a state document. Direct your technology and procurement leads to convert that list of product names into a register that records, for each product, what data it touches, what AI capabilities it currently has, and who at the vendor must notify you before those capabilities change. Then add a contract clause requiring written notice before any new generative or agentic feature is enabled in an existing product, with a right to disable it. This is the same failure Fulton County acknowledged in its Policy IFBI in August when it conceded AI is embedded in tools it has already approved. Kentucky has now given you the language to raise it without being the first district in your state to say it out loud.

6. A Peer-Reviewed Scoping Review of 22 Studies Sorts K-12 Generative AI Risk Into Three Domains, and Cognitive Dependency Sits at the Center

Source type. Peer-reviewed journal article, scoping review, gold open access. Published online February 25, 2026; June 2026 issue.

Tao, S., Lan, M., Wang, M., & Li, H. (2026). Potential risks of generative artificial intelligence integration into K-12 education: A scoping review. Computers and Education: Artificial Intelligence, 10, 100561. doi.org

The review synthesizes 22 empirical studies from K-12 contexts and organizes documented risks into three domains. The first is risks to psychological well-being, with evidence of emotional disconnection and social isolation. The second is risks to intellectual agency, comprising cognitive dependency, distorted self-assessment, and the erosion of creative authorship. The third is risks to ecological environments, including limited institutional readiness, unclear governance, equity gaps, and privacy concerns. The authors identify mitigation strategies: designing tasks that value process over product, using generative AI to provide scaffolding, such as hints, rather than direct solutions, and embedding tools within critical AI literacy curricula. The review calls for developmentally responsive governance and developmentally informed future studies. The article is published as gold open access under a CC BY-NC-ND 4.0 license and was funded by the Education University of Hong Kong. Read next to this week's IBM survey, one number connects them: 52 percent of classroom teachers name student dependency on AI as a leading concern, which is the practitioner-reported version of the intellectual agency domain this review documents in the literature.

Leadership implication. This is the citation to attach to your board memo if you restrict, and it is also the citation that limits what you may claim. It supports the statement that documented harms cluster in well-being, intellectual agency, and institutional readiness. It does not support the claim that AI harms learning at any specific grade level, because no one has measured it. Give your communications team both sentences. The three mitigation strategies are more immediately actionable than the restriction debate: instruct your curriculum office to write process-over-product task design and hint-based, rather than answer-based, AI use into the assignment guidelines that your teachers actually receive this fall.

7. A Null Result: Twenty-Two Elementary Students, Five Days With an AI Tutor, No Significant Change in Interest and a Contradicted Comprehension Claim

Source type. Preprint, not peer-reviewed. Doctoral dissertation, University of Central Florida, 2024, posted to arXiv August 5, 2026.

Holman, K. (2026). Exploring fraction comprehension and interest in elementary education through AI-powered personalized learning (arXiv:2608.04892) [Preprint]. arXiv. arxiv.org

The study is a quasi-experimental, mixed-design comparison of Mathbot, an adaptive web-based chatbot built on ChatGPT 4-Turbo and delivered through chatbotkit.com, against business-as-usual classroom instruction. The analyzed sample comprised 22 fourth- and fifth-grade students drawn from one fourth-grade and one fifth-grade classroom in an inclusive suburban school in the southeastern United States. The intervention ran over five consecutive school days in 40-minute sessions during regular mathematics periods. Repeated-measures ANOVA was used to assess changes in fraction comprehension and situational interest. On situational interest, the finding is consistent across all versions of the document: no statistically significant improvement for either group. On fraction comprehension, the document contradicts itself. The front matter reports modest improvement for Mathbot users compared with traditional instruction. The chapter-level summary states that both groups gained and that the business-as-usual group slightly outperformed Mathbot users at post-test. The author's stated conclusion is that automated personalization did not displace the teacher's instructional role and that teacher decision-making remained central to student outcomes.

Leadership implication. Use this as a procurement training artifact, not as a finding. Hand the abstract and the chapter summary side by side to whoever evaluates your instructional software and ask which sentence would have reached your board. The abstract-level claim is the one a vendor deck would quote. Then write into your evaluation rubric that any efficacy claim submitted by a vendor must be accompanied by the full results section and the sample size on the same page, and score a claim that arrives without them as incomplete rather than unsupported.

Emerging Strategic Themes

Theme 1. Literacy, access, use, and autonomy are four different decisions. This week, they collapsed in public, and the confusion will reach your board. A student can learn how AI works without receiving unrestricted access to it. A teacher can use AI without delegating a decision to it. A student can use a tool without permitting an agent to complete work autonomously. New York City deliberately separated literacy from access, teaching all high schoolers while restricting access for nearly 600,000 younger students. Write those four terms into your policy as four separate permissions, each with its own approval threshold, because a single yes or no cannot carry them.

Theme 2. The instrument matters as much as the decision. New York City and Los Angeles reached comparable restrictions within a day of each other. One did it through a published policy with pilots, exceptions, a coalition, and an evaluation commitment. The other did it through a filtering category, and its board chair learned about it in public. Both are defensible outcomes. Only one produces a record a successor can review, a parent can appeal, or a board can revise. If your district cannot name which technology decisions require a vote, you do not have a governance model; you have a console.

Theme 3. AI literacy is being scheduled, not designed. New York City will give every high school student two 45-minute modules a year. The peer-reviewed classroom evidence points elsewhere. Higgs, Kaimana, Wilgus, and Isero, publishing in the Journal of the Learning Sciences on August 4, 2026, followed high school students discussing AI across a full English language arts unit and documented reasoning moving from personal concern to systemic critique of algorithmic bias and surveillance through sustained small-group argument, not content delivery. Three of the four authors are practicing high school teachers. Ninety minutes a year is a scheduling decision. Assign AI literacy to English language arts and social studies chairs with release time for unit design, and write teacher facilitation into professional development before you write seat time into the handbook.

Theme 4. Governance has to follow the capability, not the product name. Kentucky said the quiet part in a state document on September 3: products your students already use daily will add AI capability on the vendor's schedule, not yours. Fulton County conceded the same thing inside Policy IFBI in August. An approved-tool list is a snapshot of a system that changes without notice. Convert the list into a register that tracks data access and capability changes, and put a notice-before-enablement clause into your next three contract renewals.

Theme 5. The evidence base is not keeping up with either direction. This week produced restrictions covering nearly a million students, two state agency documents, a national survey of 2,048 adults, a scoping review of 22 studies, and an elementary trial of 22 children. Restraint is now being written on the same thin record that adoption was. That is not an argument against restraint. It is an argument for building the evaluation apparatus into the restriction, which New York City did through its coalition and its promised report, and which most districts copying the restriction will not.

What Was Not Found

No peer-reviewed K-12 AI outcome study published inside the fourteen-day research window met the selection standard. Searches of academic indexes and journal listings for August 24 through September 6, 2026 returned one in-window K-12 item, a conceptual preprint on trust calibration posted September 2, with no sample and no outcome data. Every research item in this edition carries a publication date outside the window and states that date in its source line. Seventeen research candidates were pulled, and sixteen were rejected on verified publication date. That is a reported result, not an omission, and it is the second consecutive week it has occurred.

No study measures the effect of a grade-band AI restriction on any student outcome. This is the week's central absence, and it now affects nearly a million children. No published research establishes what happens to reading, mathematics, engagement, or teacher workload when a district restricts generative AI by grade. New York City has assigned that question to a coalition it created the same day. Los Angeles has assigned it to a committee that will not report until the end of the school year. Until those exist, districts copying the policy are copying an untested instrument, and districts refusing it are refusing on equally thin grounds.

No evidence separates students with disabilities or multilingual learners inside a restriction. New York City carves out assistive technology, students with disabilities, and multilingual learners, correctly and without evidence. Available reporting does not establish whether the Los Angeles filter-level block carries equivalent exceptions, which is a materially different question when the control is a category in filtering software rather than a clause in a policy. No study measures whether a restriction with carve-outs helps or harms the carved-out population, or whether the carve-out becomes a visible marker inside a classroom.

No data exists on what an AI literacy module accomplishes at ninety minutes a year. New York City will deliver two 45-minute modules to every high school student. The peer-reviewed classroom evidence describes a mechanism operating through weeks of small-group subject-area discussion. Nobody has measured the dose-response relationship between AI literacy seat time and any downstream capability, so the ninety-minute figure is a scheduling decision presented as an instructional one.

No state or district has published the rubric that defines mission-critical to learning. The most consequential sentence in the New York City announcement rests on an undefined term. Searches for a published review instrument, scoring criteria, or appeal process returned no results. Districts adopting the same language this fall will be adopting a standard that its author has not yet written down.

No one has measured whether training closes the gap the IBM survey documents. Twenty percent of educators report extensive AI training, compared with the majority using AI weekly in the classroom. That is a well-established descriptive gap. What no study establishes is whether training changes teacher AI decisions, student outcomes, or policy compliance, or how much training is enough. Districts are about to spend professional development money based on a survey that measures the problem rather than the remedy.

No elementary literacy evidence appeared anywhere in this window. Nothing located examines the use of AI in early reading or writing instruction. The restrictions announced this week fall hardest on exactly those grades. The evidence base is thinnest where the youngest children are, which has now been true in this brief for five consecutive months.

The pattern is unchanged, and the direction is new. Mandates outran evidence for three years while districts adopted. This week, mandates outran evidence while two districts withdrew. The correct response in both cases is the same: monitor what you decided, write down why you decided it, and build the exit ramp before you need it.

Novo Executive Summary

In a single week the two largest school systems in the United States restricted student-facing generative AI for close to a million children, a state education agency published a prohibited-use list naming surveillance, discipline, teacher evaluation, and IEP eligibility, another state told districts that the products they already own will become AI products without asking, and a national survey found weekly classroom AI use in the majority of secondary classrooms against extensive training in one in five. The peer-reviewed record available to justify any of it consists of a scoping review of 22 studies and a classroom trial involving 22 children. The governance question has inverted. For three years the risk was adopting without architecture. The risk now is that a restriction without inventory, a rubric defining what survives review, named decision rights, and an attached evaluation instrument is a filter setting rather than a policy, as Los Angeles demonstrated this week. The differentiator is not whether a district permits or prohibits. It is whether decision rights are written down, whether procurement specifications require something a vendor cannot self-assert, whether an evaluation cadence survives a change in leadership, and whether literacy pathways are role-based rather than generic. That architecture is what Novo Innovative Pathways builds with district leaders, and it is what makes a restriction reversible and a permission defensible.

Watch This Week

• Tuesday and Wednesday, September 8 and 9, 2026. The Kansas State Board of Education is scheduled to consider a draft statewide policy on the effective use of instructional technology at its monthly meeting in Topeka. Per the Kansas State Department of Education's announcement on September 2, 2026, the draft would require districts to address the purpose and intended use of technology; student access to and use of technology; digital citizenship and student well-being; artificial intelligence; privacy and security; and technology-use expectations and violations. It is a discussion draft, not a vote, and no district compliance date has been published.
• Wednesday, September 16, 2026. The Florida State Board of Education is scheduled to vote on the artificial intelligence amendment to Internet Safety Policy Rule 6A-1.0957. Notice of Proposed Rule 31301432 was published August 26, 2026, in Florida Administrative Register Volume 52, Number 166, opening a 21-day comment period running August 26 through September 16. The rule would require every district school board and charter school governing board in Florida to adopt and implement an amendment to its internet safety policy covering artificial intelligence. This resolves a carry-forward item that has been open since Edition 27.
• Wednesday, October 21, 2026. The next meeting of the Los Angeles Unified Generative AI Ad Hoc Committee. Subsequent dates reported are December 9, 2026, and January 20, March 3, and April 7, 2027, with policy recommendations due to the full board by the end of the 2026-27 school year. Watch whether the current filter-level restriction is ratified as board policy, and whether the committee publishes an exceptions standard for students with disabilities and multilingual learners.
• Resolved and carried forward from Editions 27 through 31. The promised New York City AI playbook arrived on September 2, 2026, as a moratorium. Watch for the Technology in Schools Coalition membership and charter and the publication date of its recommendations report, none of which appeared in the announcement.
• Carried forward from Editions 29 through 31. California adjourned sine die August 31, 2026. AB 2392, the generative AI procurement standards bill for the community colleges and California State University, and AB 2656, the 45-day labor notice bill reaching local public employers, are both with Governor Newsom. AB 1159, extending student privacy protections to digital operators that market to schools, was concurred in and enrolled on August 31. No signature or veto on any of the three was confirmed as of September 6.
• Carried forward from Editions 30 and 31. New York Governor Kathy Hochul has taken no action on S9051, the AI companion chatbot safety bill, or A6578, the AI Training Data Transparency Act. Her deadline is December 31, 2026. On September 3, 2026, legislators publicly urged action. Governor Hochul is quoted in the New York City announcement commending the moratorium. New York Senate Bill S10685, the FOCUS Act introduced by Senator Andrew Gounardes on August 21, 2026, remains referred to the Committee on Rules; Senator Gounardes is also quoted in the New York City announcement.
• Friday, September 4, 2026. Representative Josh Gottheimer announced the AI LABS Act. This proposed federal grant program would direct the Department of Education to fund K-12 AI labs and teacher training, with a mandated study of how the funds are used. No bill number was located as of September 6. Announced, not introduced, and not law.
• Fulton County Schools, Georgia. Board Policy IFBI remains unconfirmed as adopted. The September 10, 2026 session is a work session, and its published bulletin contains no minutes approval and no AI item. The September 17, 2026 board meeting is the likely venue for approval of the August 20 minutes. Note that Fulton County has migrated its board records from BoardDocs to Simbli and states that the old BoardDocs links carry outdated information. Also watch the Atlanta Board of Education, whose Policy Review Committee discussed a draft AI use policy the week of August 9, 2026, and whose first read has not been confirmed by the full board.
• Still unresolved. The GAO study of AI in K-12, requested by a Senate letter dated June 4, 2026, has no confirmed start date, engagement number, or expected publication date. North Carolina H301 has not taken any action since the conference committee appointment in June. Michigan SB 760 remains with the House Committee on Communications and Technology. The October publication of the Nagashima and colleagues CSCW study of teacher and student misalignment over classroom AI control, first flagged in Edition 26, is still pending.

Sources

Governance and Policy

Florida Department of State, Division of Library and Information Services. (2026, August 26). Notice of Proposed Rule 31301432, Rule 6A-1.0957, Internet Safety Policy. Florida Administrative Register, 52(166). flrules.org

Kansas State Department of Education. (2026, September 2). State Board to consider draft policy for instructional technology [News release]. ksde.gov

Kentucky Department of Education, Office of Education Technology. (2026, September 3). AI in Kentucky K-12, September 2026. education.ky.gov

Merod, A. (2026, September 3). LAUSD restricts all students from using AI tools. K-12 Dive. k12dive.com

Dale, M. (2026, September 2; updated September 3). LAUSD students barred from using AI, to the surprise of the school board and parents. LAist. laist.com

Office of the Mayor of the City of New York. (2026, September 2). Mayor Mamdani and Chancellor Samuels put students first with nation's broadest generative AI moratorium in schools [Press release]. nyc.gov

Office of the State Superintendent of Education. (2026, September 1). OSSE releases AI model policy to guide responsible staff use in schools [Press release]. Government of the District of Columbia. osse.dc.gov

Research, Peer-Reviewed

Higgs, J. M., Kaimana, M., Wilgus, W., & Isero, M. (2026). Leveraging onto-epistemic heterogeneity for outward exploration: Cultivating critical AI awareness in the ELA classroom. Journal of the Learning Sciences. Advance online publication. doi.org

Tao, S., Lan, M., Wang, M., & Li, H. (2026). Potential risks of generative artificial intelligence integration into K-12 education: A scoping review. Computers and Education: Artificial Intelligence, 10, 100561. doi.org

Research, Preprint, Not Peer-Reviewed

Holman, K. (2026). Exploring fraction comprehension and interest in elementary education through AI-powered personalized learning (arXiv:2608.04892) [Preprint]. arXiv. arxiv.org

Institutional Report, Not Peer-Reviewed

IBM. (2026, September 2). New IBM study finds AI adoption is outpacing K-12 readiness [Press release]. IBM Newsroom. newsroom.ibm.com

IBM. (2026). AI readiness in U.S. schools [Report]. newsroom.ibm.com