Florida Asks a Court to Bar ChatGPT From Minors, and Your District Is the Customer of a Defendant

Framing

Four editions ago, this brief began sorting governance by instrument. Edition 32 recorded a moratorium and a stoplight policy. Edition 33 recorded a statute and a union-negotiated vendor memorandum. Edition 34 recorded Florida amending a rule every district already maintains, with an opt-in consent default and an evidence review that missing studies cannot block on their own. Edition 35 recorded Maryland's statutory clock running out. This week adds an instrument no earlier edition had to name: the lawsuit. On September 28, twelve days after the Florida State Board of Education adopted its rule, the state's Attorney General asked a Highlands County court for a temporary injunction barring OpenAI from offering ChatGPT to minors in Florida. One state is now writing the terms on which its districts may use generative AI and, in a separate arm of government, asking a judge whether one of the most widely used products should be offered to children at all.

Read the instrument, not the headline. A rule binds districts. A statute binds districts and agencies. A motion for a temporary injunction binds no one until a judge signs it, and it targets the vendor rather than the school. That is why it travels. The Attorney General needs no district's cooperation, no rulemaking calendar, and no legislature in session. According to local reporting, the underlying suit was filed June 1; OpenAI removed it to federal court, and a federal judge rejected the jurisdictional arguments and returned it to Highlands County. None of the six items of relief name a school. Your district is not a party. It is a customer, and a customer of a defendant inherits the defendant's exposure without a seat at the hearing. California showed the same asymmetry from the other direction on September 30. The Governor signed a procurement template for public colleges and a records rule for AI agents. He vetoed the bill that would have fixed a 45-day notice clock before a public employer deploys generative AI.

The research counterpoint is quiet, and it points at the same gap. Lamb and Wentzell found that Connecticut districts put AI in different places in their own documents, in disciplinary sections in some and in instructional guidance in others, which means the office that owns the policy depends on the district. Kilinc and colleagues found that one chatbot, in one school, produced a pattern of coaching, modeling, and reflection in some student groups and off-task exchanges and cognitive offloading in others. Together they describe the capacity failure. When a court or a state acts on a product, a district must know where its AI policy lives, who owns it, and which instructional designs depend on the product being restricted. Most districts still cannot answer all three, and no contract currently specifies the instructional design that appears to separate good outcomes from poor ones.

The timing argument is the renewal calendar. Florida districts have until July 1, 2027, to finalize policies under the rule adopted September 16, and the Highlands County case will move on its own schedule within that window. Districts in every other state can read the filing today, because a filing is public the day it is made. The contract language that protects you from an injunction, an order, or a recall costs a conversation to request before you sign. After a court acts, it costs a negotiation. Add the clause at the next renewal, and inventory this week which of your student-facing tools run on the defendant's models.

Top Research and Policy Signals

1. Florida's Attorney General Asked a Court to Bar OpenAI From Offering ChatGPT to Minors in Florida, and the Relief Requested Does Not Mention a Single District

Source type. Pending motion in a state civil case, filed by the Florida Attorney General. September 28, 2026. An allegation and a request, not a ruling and not law. Not research.

Office of the Attorney General, State of Florida, Department of Legal Affairs. (2026, September 28). Plaintiff's motion for temporary injunction, Office of the Attorney General, State of Florida, Department of Legal Affairs v. OpenAI Global, LLC, et al., Case No. 26000295GCAXMX (Fla. 10th Jud. Cir. Ct., Highlands County). myfloridalegal.com (motion) Press release: myfloridalegal.com (release)

Florida Attorney General James Uthmeier filed the motion on September 28, 2026, in the Circuit Court of the Tenth Judicial Circuit in Highlands County. The caption names OpenAI Global, LLC and others. The Attorney General's press release confirms the filing and links to the motion but provides no details.

The motion requests an order directing the defendants to take six actions. First, to cease developing AI models without independent third-party guardrails and approval. Second, to stop offering ChatGPT to minors in Florida. Third, to cease collecting data from children under 13 without parental consent, including through age verification and parental notification procedures. Fourth, to cease misrepresenting ChatGPT's safety, reliability, and accuracy. Fifth, to stop anthropomorphizing the product through first-person language or suggestions of human attributes. Sixth, to eliminate conversation-prolongation engagement tactics.

Local reporting by News4Jax describes a 38-page motion, an underlying lawsuit filed June 1, OpenAI's removal to federal court, and a federal judge's rejection of the defendants' jurisdictional arguments that returned the case to Highlands County. The same report lists the claims as violations of the Florida Deceptive and Unfair Trade Practices Act, negligence, gross negligence, defective design, failure to warn, fraudulent misrepresentation, and public nuisance, and states that the filing includes no defendant response or hearing date and that discovery had not begun.

Study context. A motion is an advocate's request, and every factual statement in it is an allegation until proven. It cannot establish that any product harms a student, that an order will issue, or what an order would require of a school-licensed product. The motion text was read through a summarizing fetch tool, so its passages are paraphrased here and not quoted.

Leadership implication. Hand your chief of technology and your general counsel the same one-page list this week: every student-facing product your district licenses that runs on OpenAI models, whether branded as ChatGPT or embedded in a third-party tool. Then add one clause to your renewal template: the vendor must notify you within a stated number of days of any injunction, order or regulatory demand that restricts service to minors, and you may suspend the product without penalty and recover prepaid fees for the suspended period. Present the list and the clause at your next cabinet meeting.

2. California Signed a Generative AI Procurement Standards Mandate for Public Colleges and a Rule Excluding AI From Public Comment, and Vetoed the 45-Day Notice Clock for Employee Organizations

Source type. Enacted state law (AB 2392, SB 1159) and a gubernatorial veto (AB 2656). Announced September 30, 2026. AB 2392 binds public postsecondary systems, not K-12. Not research.

Governor of California. (2026, September 30). California's nation-leading AI framework just got stronger: Governor Newsom signs more first-in-the-nation worker protections and more [Press release]. gov.ca.gov Governor of California. (2026, September 30). Governor Newsom issues legislative update 9.30.2026 [Press release]. gov.ca.gov

The Governor's September 30 AI release lists two bills as signed. Assemblymember Mike Fong's AB 2392 is titled Public postsecondary education: generative artificial intelligence systems: procurement standards: training. Senator Christopher Cabaldon's SB 1159 is titled Artificial intelligence: transparency and governance. The Governor's September 30 legislative update lists AB 2656 by Assemblymember Cottie Petrie-Norris, Public employees: notice: artificial intelligence performing service within scope of work, as vetoed.

The operative text of the two signed bills was verified in Editions 31 and 33. AB 2392 requires the California Community Colleges and the California State University. It requests the University of California to convene a joint working group and deliver generative AI procurement standards and training recommendations by January 1, 2028, with product-specific training, annual review, and completion records. The Assembly passed it 77 to 0, the Senate 38 to 0, and concurrence 79 to 0. SB 1159 adds Government Code Section 54951.5, which excludes AI systems, autonomous agents, and robots from the terms person, interested person, participant, and member of the public under the Brown Act, the California Public Records Act, and several other statutes. The Assembly passed it 74 to 0, and the Senate concurred 37 to 0.

AB 2656 would have required a public employer to give recognized employee organizations 45 days' written notice before developing, buying, or requiring generative AI to perform work within represented job classes. The veto message PDF has no machine-readable text, so the Governor's reasons were not read and are not characterized.

Study context. Statutes and a veto, not studies. The vote counts come from the legislative record as verified in earlier editions and were not reopened this week; the signing and veto are confirmed on the Governor's pages. These actions don't provide evidence that generative AI procurement standards improve outcomes. AB 2392 does not reach K-12.

Leadership implication. Give your board clerk and counsel SB 1159 and ask one question before your next public meeting: does your public comment and records request procedure define who counts as a person, and what happens when an AI agent generates a comment or request? Give your human resources director the vetoed AB 2656 and ask for a written answer on what notice your bargaining agreements already require before you deploy generative AI in represented work. Then copy AB 2392's structure: a register of every generative AI contract, the training each product requires, and a completion record.

3. A Statewide Analysis of Connecticut District Policies Found AI Filed in Disciplinary Sections in Some Districts and Instructional Guidance in Others, With Definitions of Misuse That Vary

Source type. Peer-reviewed journal article. Educational Policy, OnlineFirst, September 29, 2026. Document analysis, Connecticut, United States.

Lamb, A. J., & Wentzell, C. R. (2026). AI policy in Connecticut: A statewide discourse analysis. Educational Policy. Advance online publication. doi.org/10.1177/08959048261490129

Alexandra J. Lamb and Charles R. Wentzell, both of the University of Connecticut, examined publicly available AI policies from Connecticut public school districts using discourse analysis and an infrastructure framework. The publisher's page lists the article as OnlineFirst on September 29, 2026.

As summarized from the page, the authors found that districts frame AI in more than one way at once: as beneficial, as an opportunity, and as potentially harmful. They found AI policy in different places within district documents, with some districts placing it in disciplinary sections and others in instructional guidelines. They found that the documentation ranged from minimal definitions and misuse policies to comprehensive guidance. The authors recommend that districts develop robust policies to guide teachers and students in using AI ethically and equitably.

Study context. A document analysis describes what districts wrote, not what happened afterward. It cannot establish whether any policy changed student or teacher behavior, whether placement in a discipline code produces different outcomes than placement in instructional guidance, or whether Connecticut is typical. The sample size could not be confirmed.

Leadership implication. Pull the three documents in which AI appears in your district today: the student handbook discipline section, the acceptable use policy, and any instructional guidance, and put them on one table at your next cabinet meeting. For each, name the office that owns it and the date it was last revised. If AI is defined differently in each, or defined only as misuse in the one parents read first, assign one owner this month to reconcile them into a single definition and a single home. Edition 34's Atlanta draft split policy from regulation; this is the same problem inside one handbook.

4. Forty-Two Eighth Graders Used One Chatbot and Sorted Into Four Profiles, and the Groups That Improved Were the Ones With a Complete Cycle of Guidance, Modeling and Reflection

Source type. Peer-reviewed journal article. Computers & Education: Artificial Intelligence, volume 11, article 100683. Available online September 22, 2026. Mixed methods, one rural school, Southeastern United States.

Kilinc, S., Aldemir, T., Misiejuk, K., Kaliisa, R., Sabanwar, V., Bicer, A., Song, D., & Yalvac, B. (2026). A mixed-methods analysis of AI scaffolding patterns and student inquiry profiles in a middle-school agriculture-STEM classroom. Computers and Education: Artificial Intelligence, 11, Article 100683. doi.org/10.1016/j.caeai.2026.100683

Selcuk Kilinc and seven colleagues studied 42 eighth-grade students working in 12 groups who used a custom AI chatbot named Aspen to refine scientific research questions and hypotheses in sustainable agriculture contexts. The design was mixed methods, with concurrent data gathering and sequential integration. The publisher's page shows the article available online September 22, 2026, with a December 2026 volume date.

The authors scored student artifacts and classified the 12 groups into four profiles: Struggling, Most Improved, Perfectionists, and High Achievers. Groups in the Most Improved profile showed strong connections between coaching, modeling, and reflection, with frequent uptake of the chatbot's models in their own discussion. Groups in the Struggling profile showed fragmented patterns that linked foundational scaffolding to off-task exchanges and cognitive offloading.

The authors conclude that effective AI support may depend not only on providing models but also on facilitating a complete pedagogical cycle that connects guidance, modeling, and metacognitive reflection. The sampling was a purposeful maximum-variation sample of 12 groups from one rural school, which the page notes may limit generalizability.

No effect sizes, test scores, or percentages are printed here. The page I could read included the sample, the profiles, and the authors' conclusion, but I could not confirm any numeric results.

Study context. Forty-two students in one school is design-stage evidence, a reason for pilot clauses and instructional protocols, not a warrant for district-wide claims. The study has no control group and cannot establish that the chatbot caused any outcome, only that the same tool coexisted with different group patterns.

Leadership implication. Rewrite the evaluation section of your next AI pilot plan to measure the instructional cycle, not the tool. Before launch, require the teacher team to document when guidance, modeling, and student reflection occur around the product, and require the vendor to describe in the contract what instructional design its product assumes. Name the curriculum director as the owner of that protocol and set a review at the end of the first grading period that compares groups, not averages. A district that buys the tool and skips the cycle is buying the Struggling profile.

Emerging Strategic Themes

Theme 1. A vendor can now be restricted by a court faster than your policy can be amended. Edition 34 recorded Florida's rule. This week, a motion in the same state targeted the vendor, not the school. A board policy takes a first read, a second read, and a calendar. An injunction takes a judge. Whether or not the motion succeeds, your contracts should assume that a product in your classrooms can be restricted from outside your building, and should say who suspends it and who pays.

Theme 2. The labor notice clock was vetoed, not settled. California fixed a procurement template for public colleges and vetoed a 45-day notice requirement for employee organizations in the same week. The bill still asks the question in every district with a bargaining agreement: when must you tell represented employees that generative AI is entering their work? Answer it in your own agreement before a state answers it for you.

Theme 3. AI policy has many homes and no one lives in all of them. Connecticut districts file AI under discipline in some places and instruction in others. Edition 35 showed Atlanta drafting a restrictive policy while a default-enrollment course ran under another department. Edition 34 showed a district split between policy and regulation. Governance fails at the seams between documents. Choose one home, one definition, and one named owner, then retire every other version in the handbook.

Theme 4. The tool is not the treatment. One chatbot in one school produced coaching and reflection in some groups and off-task exchanges and offloading in others. Edition 34's Instruction Partners review reached a similar worry about chatbot access during assignments. Procurement language that names the product without naming the instructional cycle around it specifies what varies least and omits what varies most across classrooms.

What Was Not Found

No court has ruled, and no hearing date or OpenAI response is stated. Florida filed the motion on September 28. The reporting I opened states no hearing date and no defendant response, and says discovery had not begun. Districts are making renewal decisions on products that may be enjoined, with no schedule that tells them when to expect an answer, and the motion as read does not address school-licensed deployments.

No outcome study of a K-12 AI product with a publisher-confirmed online date inside September 21 through October 4 was located. I searched the Consensus index with two anchor queries and surveillance, bias and equity, disability, and over-reliance queries. I swept Educational Policy, Computers and Education, Computers and Education: Artificial Intelligence, the Journal of Computer Assisted Learning, Education and Information Technologies, and several others through publisher and indexing records. The in-window K-12 items were a document analysis, a 42-student classroom study with no control group, and a cross-sectional survey of 3,629 teachers and students in Serbia, where 93.4 percent of students and 72.6 percent of teachers reported using generative AI. That survey measured self-reported perceptions, not learning outcomes.

No in-window result reports effects for English learners or students with disabilities. None of the in-window records I read disaggregates results for English learners or students with disabilities. I ran a Consensus query on AI accommodations and IEP use. I found a conceptual framework for using generative AI to monitor IEP goals, which proposes a practice and reports no data. Districts are deploying tools to these populations, and the Florida rule carries exceptions for them, without a finding.

No peer-reviewed, in-window study of classroom surveillance, engagement detection, proctoring or AI writing detectors in K-12 was located, and no new review of any AI product. The counter-evidence sweep covered the surveillance and privacy query, engineering and indexing records, and the evidence clearinghouses. I didn't find any dated items inside the window, and I couldn't locate any new What Works Clearinghouse, Evidence for ESSA, or Regional Educational Laboratory reviews of an AI product. Edition 34 recorded that of 16 fully profiled products, seven had independent studies of achievement; no update to that count appeared.

No federal action on AI in schools dated inside the window was located, and no GAO report, hearing or Federal Trade Commission action on children and AI. The Federal Register search found no Department of Education AI notice dated September 28 through October 2; the Department's press page returned an access error; the Federal Trade Commission press-release listing for September 24 through October 4 showed nothing on children, COPPA, or AI, and the only GAO item located is the June letter requesting a study. Federal guidance remains the August 20 Dear Colleague letter.

No Georgia district AI adoption inside the window was located, but Lane 7 coverage was partial. Atlanta's draft Policy IFBI remains advanced for further review, with the next committee date on the calendar as October 21 and no agenda confirmed. Fulton's district bulletin states IFBI was approved on August 20. DeKalb's page for its Gemini certification program states release on September 21 and says no action is required of parents or students; I found no board action or data privacy terms. Cherokee, Henry, Savannah-Chatham, Cobb, Clayton, and Forsyth produced nothing dated in the window; Gwinnett's board portal was unreadable; and I did not search the Tier 2 districts.

The pattern is that enforcement is outrunning evidence on two fronts. A state is asking a judge whether a product should reach minors while the research that would inform the judge's question (outcomes by age, by subgroup, and by instructional design) is absent from the week's literature. A district cannot wait for that record, because a motion, a rule, and a veto each moved this week without it. The correct response is monitoring, documentation, and exit ramps: a list of the tools you run and what they are built on, a contract that lets you suspend a product the day an order issues, and a pilot design that measures the instructional cycle. The gap exists now, as districts renew.

Novo Executive Summary

On September 28, Florida's Attorney General asked a Highlands County court to bar ChatGPT from minors in the state, twelve days after Florida's State Board adopted the rule that tells districts how to use generative AI, and two days before California's Governor signed a procurement-standards mandate for public colleges, a records rule excluding AI from public comment, and vetoed a 45-day labor notice clock. The week's research is small and exact: Connecticut districts file AI policy in more than one place, and 42 eighth graders showed that one chatbot can produce opposite patterns depending on the instructional cycle around it. The through line is not tool selection. It is architecture: who holds decision rights to suspend a product, what a procurement specification requires about continuity and instructional design, how evidence is defined and by whom, and which single document a parent, a teacher, and a judge would all find first. Novo Innovative Pathways builds that architecture with district leaders, domain by domain, so that when a court, a state, or a vendor moves, the district already knows what to do and who does it.

Watch This Week

• Circuit Court of the Tenth Judicial Circuit, Highlands County, Florida, Case No. 26000295GCAXMX. Watch for a hearing date on the temporary injunction motion filed September 28, OpenAI's response, and any order. No date is stated in the reporting opened this week. Discovery had not begun.
• Wednesday, October 14, 2026. The Kansas State Board of Education meeting packet for October 13 and 14 lists Item 11, Receive Policy Regarding the Effective Use of Instructional Technology in Kansas Schools, as a receive item, not a vote. The September minutes plan a final vote in October or November. The draft names AI as one of the topics local plans must address.
• Atlanta Board of Education. Monthly Board Meeting Monday, October 5, and Policy Committee Wednesday, October 21, 12:15 to 1:45 p.m., then Monthly Board Meeting Monday, November 2, per the board's meeting calendar. These are calendar entries, not agendas. Draft Policy IFBI was advanced for further review on September 16, with news reports pointing to November. Watch Simbli for any agenda that carries it.
• Wednesday, October 21, 2026. The Los Angeles Unified Generative AI Ad Hoc Committee meets next, per LAist and EdWeek reporting from early September; I located no LAUSD board document for the date.
• Sunday, November 1, 2026. The date Microsoft states its National AI Safety and Privacy Standard protections reach every school district it serves. OpenAI and Anthropic said by mid-September that they were working with the AFT on their own agreements; I found no signed agreement from either. Report this as no signature found, not as a refusal.
• Tuesday and Wednesday, November 10 and 11, 2026. Paris Peace Forum, where the Coalition for Children's Rights and Protection in the Age of AI states it will present its first voluntary commitments.
• Florida Department of State and Maryland State Department of Education. The Florida Administrative Register history for Rule 6A-1.0957 shows a Notice of Change on September 18, 2026, Vol. 52/182, as its last entry and no adoption or effective date; the State Board adopted the rule September 16. Maryland has still published no release date for its June 2026 guidance. Resolved this week: California's September 30 deadline, with AB 2392 and SB 1159 signed and AB 2656 vetoed; and Fulton County's August 21 bulletin, which states Policy IFBI was approved August 20. Still unresolved: DeKalb's Gemini certification program board and privacy terms, and Governor Hochul's action on S9051 by December 31.

Sources

Governance and Policy

Atlanta Public Schools. (2026). Board of Education meetings calendar. atlantapublicschools.us

Fulton County Schools. (2026, August 21). Board bulletin for 8/21/2026. news.fultonschools.org

Florida Department of State. (2026). Rule 6A-1.0957, Internet Safety Policy, rule history. Florida Administrative Rules, Register and Code. flrules.org

Governor of California. (2026, September 30). California's nation-leading AI framework just got stronger: Governor Newsom signs more first-in-the-nation worker protections and more [Press release]. gov.ca.gov

Governor of California. (2026, September 30). Governor Newsom issues legislative update 9.30.2026 [Press release]. gov.ca.gov The veto message PDF, VETO-msg-AB-2656.pdf, loaded without machine-readable text and was not read.

Kansas State Department of Education. (2026). Packet of board materials for October 13 and 14, 2026. ksde.gov

Office of the Attorney General, State of Florida, Department of Legal Affairs. (2026, September 28). Attorney General James Uthmeier files temporary injunction against OpenAI and its CEO [Press release]. myfloridalegal.com

Office of the Attorney General, State of Florida, Department of Legal Affairs. (2026, September 28). Plaintiff's motion for temporary injunction, Office of the Attorney General, State of Florida, Department of Legal Affairs v. OpenAI Global, LLC, et al., Case No. 26000295GCAXMX (Fla. 10th Jud. Cir. Ct., Highlands County). myfloridalegal.com

News4Jax. (2026, September 28). Uthmeier asks court to temporarily block OpenAI from offering ChatGPT to minors. news4jax.com

Research, Peer-Reviewed

Kilinc, S., Aldemir, T., Misiejuk, K., Kaliisa, R., Sabanwar, V., Bicer, A., Song, D., & Yalvac, B. (2026). A mixed-methods analysis of AI scaffolding patterns and student inquiry profiles in a middle-school agriculture-STEM classroom. Computers & Education: Artificial Intelligence, 11, Article 100683. doi.org

Lamb, A. J., & Wentzell, C. R. (2026). AI policy in Connecticut: A statewide discourse analysis. Educational Policy. Advance online publication. doi.org

Djukic Popovic, S., Viduka, D., Zunic, D., Stokic, A., & Popovic, S. (2026). Generative artificial intelligence in K-12 education: Teachers' and students' perspectives from a large-scale study in Serbia. AI, 7(10), Article 390. doi.org

Research, Preprint (Not Peer-Reviewed)

Nothing qualified in this category this week. Preprint sweeps located two K-12 postings dated September 21 through 25, one on children's design of trustworthy chatbots and one on three elementary teachers in a summer program, but the arXiv pages could not be read directly, and the second has a sample of three, so neither met the selection standard. They are not carried as findings.

Institutional Report (Not Peer-Reviewed)

Nothing qualified in this category this week. The editor located a Center for Democracy and Technology school leader toolkit dated September 24 and a September 28 public radio story on districts buying AI with little evidence, but did not open either report; neither includes outcome data, so neither is cited.