Florida Makes Parents Opt In to AI and Lets Tools Pass Without Evidence

Framing

This brief has followed the Florida rule since Edition 27, when it was a July draft and the ed-tech trade association filed comments arguing that opt-in consent would cut struggling readers, English learners, and students with disabilities off from adaptive tools. Edition 32 recorded the August 26 notice and the September 16 vote date. In between, the governance chain this brief has been documenting added a statutory no-training rule in California, a union-negotiated vendor standard from Microsoft, a moratorium in New York City, an administrative block in Los Angeles, and a board policy in Fulton County. Every one of those instruments answered a narrower question than Florida just answered. Florida did not decide whether students may use AI. It decided who consents, what the district must publish, what records the tool must keep, what the state must be told, and what the district must review before it renews.

Read the mechanism, not the headline. The State Board chose an amendment to an existing rule, the internet safety policy every district and charter already maintains under Section 1006.1494, rather than a new statute or a standalone AI rule. That choice matters for two reasons. First, it travels: the compliance vehicle already exists in every Florida district, so the AI section edits a document the board has already adopted, not a new program with a new owner. Second, it narrowly defines the covered object. An AI instructional tool under the rule is a generative AI application made available to a student for instruction, tutoring, practice, feedback, or assignments. The definition expressly excludes AI systems that do not interact with students, or that analyze, classify, predict, detect, or optimize decisions from existing data. The monitoring, filtering, early-warning, and proctoring systems that this brief's counter-evidence lane has repeatedly documented as the least-evidenced category in the market sit outside the rule's front door. The prohibition on undisclosed behavior monitoring, social scoring, and psychological profiling applies to AI instructional tools. It does not reach systems built for those exact purposes.

The research counterpoint is about the review the rule now requires. Element 10 directs each board to review the effectiveness of AI instructional tools, including evidence of improving student outcomes such as ESSA studies or other independent studies. It requires vendors to disclose known instructional limitations and any available independent evaluations. Then it adds the sentence that will do the most work in practice: the absence of such evaluations or studies may not, by itself, preclude approval. Two days before the vote, Instruction Partners published the closest thing the field has to an inventory of that shelf. Of sixteen student-facing products with complete profiles, seven have independent reviews of student achievement across student groups. Four are still running external studies. Three have only internal data. The Florida review will be conducted, in most districts and for most tools, against an empty column, and the rule permits approval anyway. That is not a criticism of the rule. It describes what the rule will produce unless the district writes down what it found.

The timing argument is a calendar with two dates. July 1, 2027 is the Florida deadline, and a district outside Florida should treat it as the date by which its own board will receive this rule from a parent, a vendor, or a legislator as the template. Use the nearer date in your renewal cycle. Every AI instructional tool contract that renews between now and next July is a chance to write the notice language, the opt-in and alternative workflow, the thirty-day interaction record, the no-training clause, and the vendor evidence disclosure into the agreement before a rule requires it. A district that waits inherits the rule's paperwork without the rule's leverage, because by then the vendor will have written its own version of the disclosure and the district will sign it rather than specify it.

Top Research and Policy Signals

1. Florida Adopted a Binding K-12 AI Rule With Opt-In Parent Consent, a Thirty-Day Interaction Record, State Reporting, and an Evidence Review That May Not Block Approval on Its Own

Source type. Adopted state rule. Florida State Board of Education, September 16, 2026. Not research.

Florida State Board of Education. (2026, September 16). Approval of amendment to Rule 6A-1.0957, Internet Safety Policy (Action Item 7) and Proposed Rule 6A-1.0957 [Rule text]. fldoe.org Agenda: fldoe.org Companion rule: Florida State Board of Education. (2026, September 16). Approval of new Rule 6A-14.0719, Artificial Intelligence (AI) Policies in Florida College System Institutions (Action Item 10). fldoe.org Press release: Florida Department of Education. (2026, September 16). Florida sets national standard for responsible AI in classrooms, from VPK through college. fldoe.org

The State Board met at Polk State College in Winter Haven and took up the amendment as Action Item 7, presented by Senior Chancellor Dr. Paul O. Burns. The rule, first adopted August 22, 2023, as an internet safety policy requirement, now carries a new purpose section, a definitions section, a fifteen-element minimum for the AI portion of each district and charter policy, and a prohibition. Each board must adopt and implement the amendment by July 1, 2027. The fifteen elements: plain-language parent notice naming the tool, the grades and subjects, the nature of student interaction, the objection process under Section 1006.28(2)(a), and, where the student will directly use the tool, the option to opt the child in with information on a substantively comparable non-AI alternative; additional review for tools used in Voluntary Prekindergarten through grade 5; use limited to supplementing teacher judgment and parental rights; compliance with Section 1000.05; application of the policy whether or not the tool was adopted as instructional material; accuracy, transparency, viewpoint neutrality, and accessibility for students and educators with disabilities; no undisclosed behavior monitoring, social scoring, or psychological profiling; an academic integrity section covering plagiarism, permissible use, verification of student-authored work, drafts, reflection on AI use, and demonstration of mastery without AI; teacher and administrator training; an effectiveness review including evidence of improving student outcomes such as ESSA studies, with vendors required to disclose known instructional limitations and any available independent evaluations, and with the absence of such studies not by itself precluding approval; a publicly accessible, regularly updated approved-tool list; interaction records kept at least thirty days for parental access under FERPA and Section 1002.22(2); a report to the Department of Education of every tool, its grades and subjects, frequency of use, and duration of student interaction, without per-student or per-teacher data; compliance with Section 1006.1494 and Rule 6A-1.09550 on online educational services; and a bar on approved tools selling, monetizing, profiling, or commercially exploiting student data to train commercial AI models, with priority to vendors that keep student data in the United States. The prohibition: no board may adopt a policy deploying AI designed, marketed, or configured to meet a student's social or emotional needs, including Social Emotional Learning, to simulate friendship, companionship, or an emotional relationship, or to employ relationship-building or anthropomorphic design features to keep a student interacting. The Department's press release adds that Commissioner Henry Mack directed updates to the K-12 computer science standards and the Florida Digital Classrooms Plan, the latter citing the August 20, 2026 federal Dear Colleague letter this brief covered in Edition 30.

Study context. A rule, not a study. State plainly what the rule does not do. Its definition of an AI instructional tool covers generative AI that interacts with students and expressly excludes systems that do not interact with students or that analyze, classify, predict, detect, or optimize decisions from existing data, so monitoring, filtering, and proctoring systems are outside its scope. And the press release's phrase "train agentic models" does not appear in the rule; the rule text says commercial AI models. The agenda and release did not state the vote count, and it is not reported here. The rule's filing and effective date had not been published in the Florida Administrative Register as of September 20.

Leadership implication. The element to hand your general counsel and your chief academic officer together is number 10, because it converts evidence review from a procurement preference into a policy requirement and then permits approval without evidence. Direct your curriculum and technology leads to produce, before your next renewal, a one-page evidence record for every student-facing AI tool in use: whether an independent study exists, who conducted it, which student groups it covered, what the vendor disclosed as known limitations, and what the district decided when the answer was none. Then have your operations lead cost the opt-in workflow, the substantively comparable non-AI alternative, and the thirty-day interaction record, because those three elements determine whether an opt-in default is administrable in your buildings or becomes a reason to stop offering the tool. Outside Florida, this is the template your board will be handed; write your version on your calendar, not on July 1, 2027.

2. An Independent Review of Twenty Student-Facing AI Products Found That Seven of the Sixteen Fully Profiled Have Independent Studies of Student Achievement Across Student Groups, and That General-Purpose Chatbots Pose the Clearest Threat to Student Thinking

Source type. Institutional report, not peer-reviewed. Observational and interview-based. Instruction Partners, released September 14, 2026.

Instruction Partners. (2026, September 14). AI in Action Learning Tour: A practical look at AI's role in instruction. ailearningtour.instructionpartners.org Findings pages: ailearningtour.instructionpartners.org and ailearningtour.instructionpartners.org

During the 2025-26 school year Instruction Partners, a Nashville-based nonprofit that supports instructional leadership in foundational reading, English language arts, and mathematics, studied twenty student-facing products that use AI to analyze student work or provide feedback to students: Amira Learning, Brisk Teaching, ChatGPT, Claude, Coursemojo, EnlightenAI, Google for Education, Goblins, Khanmigo, Atlas by Kiddom, Kira Learning, LitLab, MagicSchool, Magma Math, Magpie Literacy, OKO, Paloma, Playlab, Quill, and Snorkl. As of June 2026, the twenty products were in use in at least 1,000 school systems. For each product, the team spoke with the product developer, educators who use it, and students using it: thirty-two classroom observations in sixteen school systems across seven states, more than fifty teacher interviews, more than one hundred student interviews, and more than twenty-five school and system leader interviews. The review sorts the twenty into three general-purpose chatbots, five education-focused multipurpose platforms, and thirteen targeted instructional tools, and reports that the chatbots pose the clearest threat to shortchanging student thinking, the platforms will produce a broad range of outcomes depending on use, and the targeted tools with a clear opinion about the teacher's role are more likely to deliver consistent learning experiences across classrooms. Of the sixteen products with complete profiles at publication, seven have independent reviews that examine student achievement across student groups in the United States, two have studies conducted outside the United States, four have external studies underway, and three look only at internal data. The review states that independently evaluated causal studies showing impact overall and for students of color, students in poverty, multilingual learners, and students with disabilities should be the expectation for all student-facing products, and that Stanford University's AI in Education Lab will conduct secondary reviews of the developer-provided evidence, with summary information on study strength and magnitude of impact due by winter 2027. Its fourth overall takeaway is stated without hedging: if students have access to general-purpose chatbots while completing assignments, student learning will be significantly damaged.

Study context. Not peer-reviewed and not causal. The design is observational and interview-based, so it can describe how a product is used and what users report but cannot establish whether any product improves or harms achievement. The review is scoped to instructional quality only and states that data privacy, AI literacy, and environmental impact are outside it. The product count on the evidence page reflects sixteen, not twenty, products. Developers supplied their own evidence, and the independent secondary review has not yet happened.

Leadership implication. Put the list of twenty products beside your own approved-tool list and mark the overlap; at 1,000 school systems, the overlap is likely. For every match, direct the team that owns Florida-style evidence review, whether or not your state requires it yet, to record which of the four evidence categories the product falls in: independent United States review across student groups, study outside the United States, external study underway, or internal data only, and to ask the vendor in writing for the study or the timeline. Then take the fourth takeaway to your academic integrity and device policy owners: the review that looked at classrooms, not surveys, concluded that unstructured chatbot access during assignments damages learning. If your district permits general-purpose chatbots for students, the burden to show a supervised design is now yours, and you should write it down before a parent quotes this report at a board meeting.

3. Thirty-Three K-12 Teachers Said They Can Teach About AI Companions and Watch for Harm in the Classroom, but the Relationships They Fear Most Form Outside Their Jurisdiction

Source type. Preprint, not peer-reviewed. Qualitative, scenario-based interview study. Posted to arXiv September 11, 2026.

Xiao, Q., Xie, W., Deng, Z., Xiao, R., Feng, Z., He, X., Zhang, S., Stamper, J., Shen, H., & Hou, X. (2026). Exploring K-12 teachers' perceptions of students' relationships with AI companions: Boundaries, intervention strategies, and design implications. arXiv. arxiv.org

The authors conducted scenario-based interviews with thirty-three United States K-12 teachers about students' relational use of AI, starting from the observation that existing AI literacy frameworks treat AI as a tool rather than a relationship and that little is known about how teachers understand or respond to students' companion use. Teachers distinguished two kinds of companions. They welcomed academic companions, systems students turn to for help with schoolwork. They worried that intimate companions remove the developmental friction through which students learn to sustain human relationships. Teachers drew the boundaries of their own jurisdiction by setting them around observable well-being: inside the classroom and where a student's well-being was visible, they taught, talked, and watched; beyond that line, they positioned themselves as the adults best placed to notice a problem and connect the student with support rather than as the adults who could intervene. They described AI companion literacy as shared work across counselors, parents, platforms, and policymakers, spiraling across grade levels rather than delivered once. The paper proposes AI companion literacy as an extension of AI literacy and discusses implications for K-12 AI education. The abstract reports no numeric results, and none are stated here.

Study context. Thirty-three teachers, one country, scenario-based interviews, and no student data. The design cannot establish how many students use AI companions, what the effects are, or whether any teacher intervention works. It establishes what a sample of teachers believe they can see and do, which is the capacity question every companion prohibition now depends on. The paper hasn't been peer-reviewed, and no venue is listed.

Leadership implication. Florida's rule, Adam's Law in California, the New York City moratorium, and the Microsoft standard all now prohibit or restrict companion-style AI, and each assumes an adult who can recognize it. These teachers say the classroom is the one place they can, and that most of the relationships they fear form elsewhere. Direct your student services lead to name, in writing, which role owns companion-use concerns when a teacher notices one: the counselor, the school psychologist, the family engagement office, or the technology office, with a referral path a teacher can follow in one step. Then have your curriculum office add the academic-versus-intimate distinction to whatever AI literacy content you already deliver, because a district that teaches prompting and citation has not touched the use these teachers are worried about. Put the referral path in the staff handbook before the spring, not in a policy nobody reads.

4. Atlanta Public Schools Advanced a Draft AI Policy That Bars AI From Grading, Promotion, Discipline, and Graduation Decisions and Bars Discipline Based on a Detection Tool Alone, With a Decision Reported for November

Source type. District board committee action, draft policy under review, Georgia. Not adopted. Not research.

Williams, V. (2026, September 16). Atlanta Public Schools moves toward adopting new AI policy. Atlanta News First. atlantanewsfirst.com Center for Civic Innovation. (2026, August 18). School board discusses draft AI use policy. civicatlanta.org

The Atlanta Board of Education's Policy Review Committee first discussed the draft in the week of August 9. It returned to it on September 15, when it advanced the proposed policy for further review by voice vote after discussion of regulations, teacher training, student due process, and public calls for an independent review board. As reported across the two meetings, the draft bars the district from using AI as a teacher or to reduce or replace certified educator positions; permits students to use AI when the use connects to a clearly identified instructional objective but prohibits using AI to complete assignments or generate their work unless an educator authorizes it; bars punishing a student for unpermitted AI use based on a detection tool alone, requiring staff review and a chance for the student to explain; bars AI from issuing grades or making decisions about promotion, discipline, or graduation; bars teachers from entering sensitive information into AI platforms while allowing planning use; requires a public list of approved tools; and requires a parent to sign off before a child uses any AI chatbot that collects data. Staff described a drafting approach in which the policy states the vision and implementation details were moved into a separate regulation. Atlanta News First reports that the district expects to make the adoption decision at a November meeting. Public commenters at the same committee cycle asked the district to put a student-facing moratorium in writing, create an independent AI review board, and publish an inventory of AI-enabled tools already embedded in district systems.

Study context. A committee-advanced draft is neither a first read by the full board nor an adoption. Provisions are as reported by two named outlets and one meeting-record service; the draft text itself was not opened. Which body takes the November decision, the full board or the administration under an existing policy, is stated differently across reports and is not resolved here.

Leadership implication. Atlanta is drafting in the same shape Fulton County adopted in August and Florida adopted this week: human decision rule, approved-tool list, parental sign-off, and now a due-process clause that the Fulton IFBI summary this brief reported in Edition 31 did not include. Copy the detection-tool clause this month. Direct your student discipline and academic integrity owners to write, in the code of conduct, that no consequence for AI use rests on a detector's output alone, that a staff member reviews the work, and that the student is heard, because Atlanta's own drafters declined to let a detector carry a consequence. Your handbook should not carry a claim the district cannot defend at a hearing. Then note what Atlanta's parents asked for and no district in this brief has yet published: an inventory of AI already embedded in the systems the district owns. Assign that inventory to your technology office with a date.

Emerging Strategic Themes

Theme 1. The consent default is now set by someone other than the district. Florida set opt-in by rule this week. Oklahoma set opt-out by statute, as Edition 27 recorded. New York City set a moratorium for grades 2-K through 8 in Edition 32. Atlanta's draft requires a parent signature before a data-collecting chatbot. Four jurisdictions, four defaults, none of them chosen by the district that will run the workflow. A district that has not written its own consent default, its alternative assignment, and its record of who opted out and why will receive one of these four from a legislature, a state board, or a parent with a printout. Choose one this fall with a written rationale, and be ready to defend it against the other three.

Theme 2. Evidence review is becoming a compliance step before the evidence exists. Florida's element 10 makes outcome-evidence review mandatory and states that the absence of studies does not, by itself, preclude approval. Instruction Partners counted seven of sixteen products with independent achievement studies across student groups. The federal Dear Colleague letter in Edition 30 told districts to buy on demonstrated outcomes with no federal shelf to buy from, and Edition 30 also reported zero What Works Clearinghouse studies on AI and student outcomes. The instrument that will matter is not the review but the record of what the review found, tool by tool, including the word none. Write that record so the district, not the vendor, holds the account of what was known at approval.

Theme 3. The definition decides what escapes the rule. Florida's AI instructional tool is generative AI that interacts with students; systems that analyze, classify, predict, detect, or optimize from existing data are excluded by name. The monitoring, filtering, and proctoring category this brief has documented as the least-evidenced in the market therefore sits outside the rule that bans undisclosed behavior monitoring. Kentucky warned in Edition 32 that the products districts already own will become AI products on the vendor's schedule, and Atlanta's parents asked this week for an inventory of embedded AI. Read every AI policy, including your own, and start with its definition, then list what it leaves out.

Theme 4. The companion prohibition has arrived faster than the adult who can see the companion. In five weeks, the companion-style prohibition appeared in the Microsoft standard, California's Adam's Law, Florida's rule, and the Florida College System definition. The one study of teacher capacity this week found teachers drawing their jurisdiction at the classroom door and describing the relationships they fear as forming beyond it. A prohibition on deploying a companion is administrable; a district can decline to buy one. Noticing the one a student already has is a different task, and no instrument enacted this season assigns it to anyone. Assign it.

What Was Not Found

No peer-reviewed K-12 AI study with a publisher-verified online date inside September 7 through 20 was located. The Consensus query bank ran both anchor queries and rotating categories 1, 5, and 6 with the year set to 2026, plus a second pass excluding preprints, plus nine journal and preprint sweeps. Twenty-six candidates were examined. Eight peer-reviewed candidates carried verified dates before the window opened, including a 371-student three-arm randomized trial of generative AI support for self-regulated learning in grades 7 through 9 in Educational Psychology Review that reported no clear condition differences on most learning outcomes; its online date could be confirmed only at rung two, and it is held for a future edition rather than run. Five were prior-edition duplicates, three had publisher records that could not be confirmed, seven were higher-education samples, and nine had no sample or design. The three candidates inside the window are all arXiv preprints, one of which runs above. The arXiv cs.CY listing for September 2026 could be read only through its first fifty of 284 entries, so the preprint sweep is incomplete and is reported as such.

No evidence exists on what an opt-in consent default does to who uses the tool. Florida's rule sets opt-in with a comparable alternative. The Software and Information Industry Association argued in July that opt-in would disproportionately remove adaptive tools from struggling readers, English learners, and students with disabilities. No study located in any window this brief has run measures opt-in versus opt-out rates by student group, the instructional quality of the alternative assignment, or the effect on achievement of either default. Florida did not commission an evaluation. Every district in the state will run the experiment without a measurement plan.

No independent outcome study exists for most of the student-facing products districts already license. Instruction Partners has seven of sixteen with independent United States reviews across student groups, four with external studies underway, and three with internal data only. The counter-evidence sweep opened vendor evidence pages for the products Florida's review will most often reach and found one, Amira Learning, with a posted Evidence for ESSA entry and stated effect sizes; other posted tiers located were for teacher-workflow outcomes rather than student learning, or could not be opened at the certifying body and are not reported. No What Works Clearinghouse addition on AI and student outcomes was located, consistent with the zero-study finding reported in Edition 30.

No federal action on AI in schools occurred in the window. The Department of Education has issued nothing on AI since the August 20 Dear Colleague letter. No Federal Trade Commission action on AI and children, no congressional hearing on AI and students, and no Government Accountability Office report dated September 14 through 20 was located. The one federal instrument found, H.R. 10315, the AI Learning, Advancement, and Building Skills Act, was introduced September 8 and referred to the Committee on Education and Workforce; it is a grant bill with one cosponsor and creates no obligation.

No Georgia board adopted an AI policy in the window, and the one adoption this brief has reported remains unconfirmed in the minutes. Atlanta advanced a draft. Fulton County's September 17 bulletin lists seven policies approved. It does not mention IFBI, EF, or EFB, and the August 20 minutes could not be read because the Simbli portal did not render. DeKalb County's Gemini AI Certification Program still has no located board action, contract, or data privacy terms two days before its release date. Gwinnett, Cobb, Cherokee, Forsyth, Clayton, Henry, and Savannah-Chatham returned no AI policy action dated September 14 through 20; this week's Tier 2 rotation of Columbia, Paulding, Houston, Fayette, and Richmond returned standing guidance and program coverage but no dated board action. The Georgia Department of Audits and Accounts' June 2026 performance audit of generative AI in K-12, with more than 13,000 teacher survey responses, is the most substantial Georgia instrument located and is outside the window; it is carried to Watch.

No instrument or study this week separates elementary literacy, English learners, or students with disabilities. Florida's additional review for VPK through grade 5 names age and developmental appropriateness and standards alignment as criteria and names no evidence standard. The accessibility element requires compliance with existing law and nothing more. The Instruction Partners review names priority groups as the expectation for evidence and reports that most products do not yet meet it. The gap where the youngest and most dependent students sit has now persisted in this brief for six months, and this week a state wrote a rule around it without closing it.

This brief has named the pattern since spring, sharpened by a rule rather than a guidance document. Mandates outran evidence while districts adopted, then while districts withdrew, and now while a state board wrote evidence review into policy and permitted approval without it. The correct response is not to wait for the studies. It is to write the evidence record for every tool now, including the entries that say none; to choose the consent default on a stated rationale; to inventory what the definition excludes; and to build the exit ramp into the contract so that when the evidence arrives, the district can act on it without renegotiating under a deadline.

Novo Executive Summary

Florida this week bound every district and charter, by state board rule, to an AI policy with an opt-in consent default, a thirty-day interaction record, state reporting, a companion prohibition, and a mandatory outcome-evidence review that the rule itself says may not block approval on its own. Two days earlier, the first independent inventory of student-facing AI products found seven of sixteen with independent achievement studies across student groups, and a preprint found thirty-three teachers drawing the boundary of what they can see at the classroom door. Atlanta advanced a draft in the same shape with a due-process clause no prior instrument in this brief's record has carried. The strategic conclusion is that the governance layer has moved from what a district may allow to what a district must be able to show: who consented, what was approved, what evidence existed, what the definition excluded, and who is responsible when a rule assumes an adult who can see. That is architecture, decision rights, procurement specification, evaluation, and literacy pathways, not tool selection. Novo Innovative Pathways builds exactly that with district leaders, from the evidence record and the consent workflow to the role-based literacy that puts a named adult behind every prohibition a state now writes.

Watch This Week

• Monday, September 21, 2026. Published release date for the DeKalb County School District Gemini AI Certification Program for middle and high school students. Watch for the enrollment process and for any board agenda item, vendor agreement, or data privacy terms; none were located as of September 20. Carried forward from Edition 33. [Flagged: rung three. The district's instructional technology page did not render for the second consecutive week; program name, grade span, and release date rest on the search index record of the district's own page.]
• Wednesday, September 30, 2026. Governor Newsom's deadline to sign or veto remaining bills. As of the Governor's September 18 legislative update, no action had been taken on AB 2392, generative AI procurement standards and training for the community colleges and California State University, presented September 3; AB 2656, the 45-day notice to a recognized employee organization before a public employer develops, purchases, or requires generative AI in represented work, presented August 31; or SB 1159, excluding AI systems, autonomous agents, and robots from the definition of a person under the state's open meeting and public records acts, presented August 28. Carried forward from Editions 29 through 33. [Flagged: rung three. The Legislature's bill pages did not render for this build; status and presentation dates were confirmed at CalMatters Digital Democracy and by their absence from the Governor's September 14 and September 18 signing lists.] Separately, Executive Order N-9-26, signed September 18, directs state recommendations by November 16, 2026 on independent verification, an on-site audit regime, and a kill switch for frontier models; its operative text contains nothing on schools or minors.
• Florida Department of State. Watch the Florida Administrative Register for the filing and effective date of the amended Rule 6A-1.0957 and new Rule 6A-14.0719, and for the Department of Education's reporting template under element 13 and any update to Rule 6A-1.09550 on online educational services. The July 1, 2027 district deadline is still in effect.
• Instruction Partners and Stanford University's AI in Education Lab. Secondary reviews of developer-supplied evidence for the twenty products are due by winter 2027; the 2026-27 tour adds at least twenty more student-facing products and teacher-facing products, with results reported in winter. A live briefing on the first-year findings is reported for Monday, September 28, 2026. [Flagged: rung three. The September 28 briefing date was reported in trade coverage and was not found on the project site.]
• Atlanta Board of Education. The Policy Review Committee's October meeting appears on the board calendar without a posted date; draft Policy IFBI returns for further review, with an adoption decision reported for November. Watch also for the district's response to public requests for an embedded-AI inventory and an independent review board. Fulton County's October board meeting should carry the September 17 minutes; the August 20 minutes, which would confirm adoption of Board Policy IFBI, were not confirmed as approved in the September 18 Board Bulletin. Gwinnett County's work session and business meeting fall on September 23 and 24. [Flagged: rung three for the Gwinnett dates, confirmed through search results rather than an opened BoardDocs item.]
• Sunday, November 1, 2026. The date Microsoft states its National AI Safety and Privacy Standard protections reach every school district it serves. As of September 20, no signature by OpenAI or Anthropic was located; report this as no additional signatory found, not as a refusal. Carried forward from Edition 33.
• Still unresolved. The Kansas State Board of Education's October meeting: no agency document confirming the November consensus target was opened this week, and the target is not restated as confirmed. Governor Hochul's action on S9051 and A6578 by December 31. The Los Angeles Unified Generative AI Ad Hoc Committee was previously reported for October 21; a second account describing a different cadence surfaced this week, and neither could be confirmed on an LAUSD board page. North Carolina H301 and Michigan SB 760, chamber status unconfirmed this week. H.R. 10315 was referred to House Education and Workforce on September 8. The Georgia Department of Audits and Accounts performance audit of generative AI in K-12 (June 2026) is for any legislative follow-up in the 2027 session. The Nagashima et al. CSCW publication, first flagged in Edition 26, is still pending for October.

Sources

Governance and Policy

Florida State Board of Education. (2026, September 16). September 16, 2026 meeting agenda. fldoe.org

Florida State Board of Education. (2026, September 16). Approval of amendment to Rule 6A-1.0957, Internet Safety Policy (Action Item 7). fldoe.org

Florida State Board of Education. (2026, September 16). Proposed Rule 6A-1.0957, Internet Safety Policy [Rule text as adopted]. fldoe.org

Florida State Board of Education. (2026, September 16). Proposed Rule 6A-14.0719, Artificial Intelligence (AI.) Policies in Florida College System Institutions [Rule text as adopted]. fldoe.org

Florida Department of Education. (2026, September 16). Florida sets national standard for responsible AI in classrooms, from VPK through college [Press release]. fldoe.org

Kissel, K. (2026, September 16). Florida approves statewide AI rules for K-12 schools and colleges. Central Florida Public Media. cfpublic.org

Williams, V. (2026, September 16). Atlanta Public Schools moves toward adopting new AI policy. Atlanta News First. atlantanewsfirst.com

Center for Civic Innovation. (2026, August 18). School board discusses draft AI use policy. civicatlanta.org

Fulton County Schools. (2026, September 18). Board Bulletin for 9/18/2026. news.fultonschools.org

Office of Governor Gavin Newsom. (2026, September 18). Executive Order N-9-26 [Signed text]. gov.ca.gov

Office of Governor Gavin Newsom. (2026, September 18). Governor Newsom issues legislative update 9.18.26. gov.ca.gov

CalMatters Digital Democracy. (2026). AB 2392, AB 2656, and SB 1159 bill status pages. calmatters.digitaldemocracy.org ; calmatters.digitaldemocracy.org ; calmatters.digitaldemocracy.org

U.S. Government Publishing Office. (2026, September 8). H.R. 10315, AI Learning, Advancement, and Building Skills Act, 119th Congress (Introduced in House). govinfo.gov

Research, Peer-Reviewed

No peer-reviewed study with a publisher-verified online date between September 7 and 20, 2026 met the selection standard.

Research, Preprint, Not Peer-Reviewed

Xiao, Q., Xie, W., Deng, Z., Xiao, R., Feng, Z., He, X., Zhang, S., Stamper, J., Shen, H., & Hou, X. (2026, September 11). Exploring K-12 teachers' perceptions of students' relationships with AI companions: Boundaries, intervention strategies, and design implications. arXiv:2609.13479. arxiv.org

Institutional Report, Not Peer-Reviewed

Instruction Partners. (2026, September 14). AI in Action Learning Tour: A practical look at AI's role in instruction. ailearningtour.instructionpartners.org

Instruction Partners. (2026, September). How products vary; Overall takeaways [Findings pages]. ailearningtour.instructionpartners.org ; ailearningtour.instructionpartners.org