Kansas Requires Every District to Adopt an Instructional Technology Policy Covering AI by July 1, 2027, and Has Not Published the Policy Text: The State Names the Topic, and the Board Writes the Clause
On November 13, 2026, the Kansas State Department of Education announced that every district and accredited nonpublic school must address artificial intelligence in an instructional technology policy by July 1, 2027. The release lists five required areas and no policy text. The same week, the AFT and UFT made their privacy standard binding on Anthropic for Claude for Teachers, and Seattle blocked Google Search AI features at every grade. This week's research counterpoint is Stanford's November evidence review, which added 32 strong causal studies and found that offloading academic tasks to AI can reduce learning.
- Governance signal. The Kansas State Board of Education approved a policy, a resolution, and Board Directive No. 2026-1 requiring every district and accredited nonpublic school to address five areas in an instructional technology policy by July 1, 2027: purpose and intended use of technology, student access and use, digital citizenship and wellbeing, artificial intelligence, and privacy and security. The release describes parent acknowledgment by 2028-29, not parent consent. Kansas superintendents now have an AI policy deadline and no published policy text to write against.
- The second axis: Anthropic is now bound by the union privacy standard, for one product. On November 10, the AFT and the United Federation of Teachers announced a legally binding agreement that applies the national AI Safety and Privacy Standard for Schools to Anthropic's existing district agreements for Claude for Teachers. Education Week reports it bars the use of student, educator, and customer data to train models, with exceptions for certain research and safety data. It covers only Claude for Teachers, not other companies' tools built on Claude. Neither source describes an audit right or a penalty.
- The third axis: the off switch is now a district operations problem. Seattle Public Schools began blocking Gemini, unapproved generative AI tools, and Google Search AI features at every grade on November 9, per KOMO. Chicago Public Schools said it is exploring all options to suppress AI Overviews and will look to other search solutions if it can't. California's state superintendent convened an AI summit on November 9, and on November 11 the California Federation of Teachers asked the governor for a three-year K-8 pause.
- Key research finding, institutional review not peer-reviewed. Stanford's SCALE Initiative published its November 2026 evidence review on November 10. It added 32 strong causal studies, 28 on student-facing AI and 4 on educator-facing AI, for 52 in total; only five new student-facing studies were set in U.S. K-12. It concludes that using AI to offload academic tasks can reduce learning and that positive student experiences don't necessarily mean stronger outcomes.
- Research counterpoint, peer-reviewed. A randomized experiment at an English sixth-form college, published online in the British Journal of Educational Technology on November 13, compared a human tutor, a generative AI tutor, and unguided internet search. Challenge-seeking didn't differ by condition. Human tutoring produced the lowest perceived difficulty and the most questions attempted. The abstract doesn't report the sample size.
- Evidence gap. No peer-reviewed K-12 trial with a standardized achievement outcome carried a verified in-window date. A peer-reviewed analysis of three student wellbeing surveillance tools, published November 6, documents what vendors claim and reports no outcome data, because none is published. I couldn't locate the text of the Kansas policy or the directive.
- Watch this week. New York City Council's letter to Google asks for answers by Thursday, November 19, and Senator Markey's by Monday, November 30. Character.AI sale objections are due Wednesday, November 18, with bids due November 24. Georgia boards meet in Forsyth on November 16, Atlanta's Policy Committee on November 18, and Gwinnett and Cherokee the same week.
Framing
Since Edition 32, this brief has tracked AI governance by instrument: a mayor's moratorium, a union and vendor standard, a state rule, a court motion, eight state lawsuits, a bankruptcy notice, a Senate letter, a statute, and last week a vendor's own default. This week brings a state board directive. Edition 35 called the compliance clock the unit of governance, and Edition 39 watched Maryland's clock run out on October 22 with several large districts holding no adopted policy. Edition 41 found that 72 percent of state guidance documents carry no version history. Kansas has now set its own clock, and it runs to July 1, 2027.
Read the mechanism. A state board can't easily write an AI rule that survives the next product release, so Kansas named the topic and left the clause to each local board. The release lists artificial intelligence as one of five required areas and adds no sub-elements; KSNT's account of the implementation plan says it must cover AI use by teachers and students, including chatbots and generative AI. The parent provision matters more than its length. The Johnson County Post reported on November 9 that the board was weighing whether parents would only acknowledge a district policy or would actively consent, and that five Johnson County districts had opposed consent in an October 16 letter. The November 13 release describes acknowledgment. Here's how I read it: an acknowledgment standard keeps instruction whole and puts the burden on the board that approves each tool to justify it. The same week, Senator Markey and the New York City Council each wrote to Google asking whether schools can keep Search while turning off the AI inside it; a Google spokesman had already told reporters that Gemini and Search are separate controls. A mandate that names AI assumes the off switch exists.
The research counterpoint is what a local board can cite when it writes that clause. Stanford's November review, which follows the March review Edition 38 cited, screened 868 papers and found 32 with strong causal evidence. Among the K-12 studies it lists, an AI tutor in grades 6 to 8 slowed practice without improving delayed learning; teacher AI access in Türkiye left average exam scores unchanged and lowered student motivation; and the clearest gains came from a Google-authored trial in Sierra Leone. The review's own takeaway is that AI can improve students' perceived learning experiences without improving learning outcomes. A Cambridge experiment published November 13 found no difference in challenge-seeking across human, AI, and search support. Districts are being asked to set policy on a record that is thin and mixed.
The timing argument is twenty months, and most of it is already spoken for. A Kansas board that adopts in June 2027 has to draft this winter, take first read in spring, and train staff before parent acknowledgment starts in 2028-29. California shows the fork in a single week: the state superintendent convened a summit on responsible adoption on November 9, and on November 11 the state's teachers' federation asked the governor for a three-year K-8 pause. Districts in other states should read Kansas as a template: name an owner for each of the five areas now, keep a dated inventory of every AI tool in use, and write the evidence question into the policy, not the handbook. If you wait for your state to publish its version, you'll write the policy under a deadline someone else chose, and you'll have less time to read what the trials actually found.
Top Research and Policy Signals
1. Kansas Will Require Every District and Accredited Nonpublic School to Adopt an Instructional Technology Policy That Addresses Artificial Intelligence by July 1, 2027, With Parent Acknowledgment Rather Than Consent
Source type. Adopted state board policy and board directive; Kansas State Department of Education news release, November 13, 2026.
Citation (APA). Kansas State Department of Education. (2026, November 13). Kansas State Board of Education requires all school districts to develop instructional technology policies; policy prioritizes high-quality instruction while preserving local decision-making [News release]. ksde.gov
The release states that the State Board approved the policy, a resolution, and Board Directive No. 2026-1 to implement it. It provides no meeting date or vote count. Districts and accredited nonpublic schools must address five areas in their instructional technology policies by July 1, 2027: purpose and intended use of technology in elementary and secondary settings; student technology access and use; digital citizenship and student wellbeing; artificial intelligence; and privacy and security. The release lists no sub-elements for artificial intelligence.
Districts must establish a process for parents to acknowledge the school's policy by the start of the 2028-29 school year. The release describes acknowledgment and doesn't mention consent or an opt-out. The policy expires July 1, 2030 unless the board acts. Commissioner Dr. Jake Steel said, "technology only belongs in the classroom when it strengthens teaching and learning."
KSNT's November 13 report adds that local boards must approve and review the policy and an implementation plan annually, post both publicly, and submit adopted plans to KSDE by July 1, 2027, and that the plan must include guidelines for AI tools including chatbots and generative AI. The Johnson County Post reported on November 9 that the board was weighing acknowledgment against active consent, which would let parents opt children in or out of school devices, and that Olathe, Shawnee Mission, Blue Valley, De Soto, and Gardner Edgerton opposed consent in an October 16 joint letter.
Study context. This release describes a policy, not the policy text. Until the text is public, districts can't know what the AI section must include, whether it carries an evidence or procurement standard, or how the directive will be enforced.
Leadership implication. If you lead a Kansas district, I'd assign one cabinet owner to each of the five areas this month and put a first read on the spring board calendar so adoption lands before July 1, 2027. Ask KSDE in writing for the policy text and Board Directive No. 2026-1 before drafting. In the AI section, list every student-facing AI tool, name who approves each one, and state what evidence it must show at renewal. Districts outside Kansas should run the same five-area gap check now, before their own state copies the template.
2. The AFT and UFT Made Their AI Safety and Privacy Standard Binding on Anthropic: It Bars Training on School Data With Research and Safety Exceptions, Covers Only Claude for Teachers, and Names No Audit Right
Source type. Union and vendor contractual agreement, not a law or a regulation; AFT press release, November 10, 2026, with reporting by Education Week and K-12 Dive.
Citation (APA). American Federation of Teachers. (2026, November 10). AFT, UFT and Anthropic sign binding AI safety and privacy agreement for schools [Press release]. aft.org
The AFT and the United Federation of Teachers announced "a legally binding agreement with Anthropic" that makes the national AI Safety and Privacy Standard for Schools enforceable for Anthropic in U.S. K-12 schools. The release says the standard automatically applies to Anthropic's existing district agreements for Claude for Teachers. Its listed terms: student and educator information used only for authorized educational purposes, never sold or used for targeted advertising; "Schools keep control over their data, including how long it is kept and when it is deleted"; educators, not AI systems, responsible for consequential decisions; and plain-language disclosure.
Education Week adds terms the release doesn't state. Student, educator, and customer data can't be used to train or improve models, except for certain research and safety-related data. Anthropic has a continuing duty to notify customers of breaches, even after a contract ends, and must minimize data collection. Higher-risk features such as biometrics, memory, profiling, and tracking require additional review. Districts must adopt policies requiring human review of AI-informed grading, discipline, or placement decisions and train staff to do it. The terms apply to existing contracts, and districts must opt in for future licensing agreements.
Claude for Teachers, per Education Week, gives verified K-12 teachers a free account with chat, agentic, and coding tools. It has no student-facing tools. The agreement doesn't necessarily cover other companies' products built on Claude models. The AFT is still negotiating with OpenAI and has no agreement with Google. Neither source describes an audit right, an enforcement mechanism, or a penalty.
Study context. The full agreement is linked from the AFT release, and districts should read it before relying on any term. The training exception, breach duty, high-risk feature review, and opt-in requirement come from Education Week's reporting and should be checked against the agreement before a district relies on them.
Leadership implication. I'd ask your counsel to pull the full agreement this month and build one comparison table across Anthropic, Microsoft, and your own contracts: covered products, training and its exceptions, deletion period, breach notice, subprocessors, audit, and who may enforce. Opt in in writing for any new Anthropic license. Note that the agreement asks districts to require human review of AI-informed grading, discipline, and placement, so your board policy must include that rule. Log every free Claude for Teachers account in your AI inventory, because a free account still moves district data.
3. Seattle Blocked Gemini and Google Search AI Features at Every Grade on November 9, and Chicago Said It Will Seek Other Search Tools if It Can't Suppress AI Overviews: The Off Switch Is Now a Procurement Dependency
Source type. District operational actions reported in news coverage (KOMO, KUOW, Chalkbeat), plus letters from Senator Markey and the New York City Council.
Citation (APA). KOMO News Staff. (2026, November 5). Seattle schools to cap elementary students' screen time, block unapproved AI tools. KOMO News. komonews.com and Seminera, M. (2026, November 12). Protest outside of Chicago Google office points to rising concerns about AI in schools. Chalkbeat Chicago. chalkbeat.org
KOMO reports that from Monday, November 9, Seattle Public Schools blocks Google Search AI features, including AI summaries, Gemini, and any generative AI tool the district hasn't approved, at every grade level, with a 60-minute daily screen cap for grades 3 to 5 starting in January 2027. KUOW reports that the district has approved and purchased MagicSchool AI and Microsoft Copilot. Board director Evan Briggs said the presentation had no data showing the tools improved reading scores: "We just don't have that data; it doesn't exist."
On November 11, about 30 people, led by Illinois Families for Public Schools, protested outside Google's West Loop office in Chicago, Chalkbeat reports. A Chicago Public Schools spokesperson said the district is "actively exploring all options to suppress" the AI overview independently, and will look to other search solutions if it can't provide Google search without it.
Google's spokesman told reporters on November 4 that Gemini and Search "are separate controls," both off by default for K-12 student accounts. The open question is the AI inside Search. By November 30, Senator Markey asked whether schools may disable AI Overviews, Gemini, and Guided Learning while keeping Search. The New York City Council asked by November 19, and its release quotes a New York Times report that Google "cannot edit its infrastructure to separate its products from AI."
Study context. These are news accounts of district actions, not adopted board policies or contracts. KOMO's report that Seattle blocks Search AI summaries sits beside Google's reported position that it can't separate them, and no source explains how Seattle does it.
Leadership implication. I'd have your chief technology officer build a feature-level register this month: for each platform students use, list every embedded AI function, the grades with access, whether it can be switched off independently, who has authority to switch it off, and the replacement if it can't. Bring the register to your board before renewal. If an essential service can't separate its AI feature, write that into your risk log and your next request for proposals as a scored requirement, not a preference.
4. California Split in One Week: The State Superintendent Convened an AI Summit on November 9, and the California Federation of Teachers Asked the Governor for a Three-Year K-8 Pause by Executive Order
Source type. State agency release (California Department of Education, News Release #26-46, November 9, 2026); and a union letter requesting an executive order, not an enacted order, reported by the California Globe (November 11).
Citation (APA). California Department of Education. (2026, November 9). State Superintendent Tony Thurmond convenes summit on AI in education, emphasizes innovation, student safety, and educator leadership (Release #26-46) [News release]. cde.ca.gov and Barth, M. (2026, November 11). California teachers union presses Newsom for three-year AI classroom pause. California Globe. californiaglobe.com
State Superintendent Tony Thurmond convened "The AI & Education Innovation Summit: Moving the Needle for California Students" on November 9. The release lists participants from CodeAI, Microsoft, Nvidia, Google, Digital Promise, Amira Learning, Carnegie Learning, Curriculum Associates, and the Center for Outcomes Based Contracting, as well as themes of literacy and math instruction, workforce preparation, privacy, safety, equity, and educator involvement. Thurmond said, "Innovation cannot come at the expense of student safety, privacy." The release says Senate Bill 1288 created a statewide AI working group to develop guidance and a model policy for local educational agencies. No funding was announced.
Two days later, the California Globe reported that the California Federation of Teachers sent outgoing Governor Gavin Newsom a letter asking for an executive order halting "student-facing technologies that include generative AI features" in kindergarten through eighth grade for three years, to give the incoming governor and legislature time to write rules. The Globe reports CFT member polling in which nearly 9 in 10 said generative AI harms critical-thinking development. The governor hadn't responded at publication.
Study context. The summit is a convening, not a rule, and the CFT request is a letter, not an order. The polling figures are the union's own member survey, not a representative sample.
Leadership implication. Whichever way your state leans, I'd write your district's decision process down before the debate reaches your board: the educational purposes that justify AI, the uses that are off limits, the evidence that permits expansion, and who signs off. If you operate in California, assign someone to track the SB 1288 working group's model policy and the governor's response. A broad ban and a broad rollout both fail the same test if no one can say how the decision was made.
5. Stanford's November Evidence Review Added 32 Strong Causal Studies, Only Five of Them Student-Facing U.S. K-12 Trials, and Finds That Using AI to Offload Academic Tasks Can Reduce Learning
Source type. Institutional evidence review, not peer-reviewed; SCALE Initiative at Stanford University, November 10, 2026.
Citation (APA). Walker, S., Fesler, L., Martinez Claeys, J. P., Agnew, C., & Loeb, S. (2026, November 10). The evidence base on AI in K-12: A November 2026 review. SCALE Initiative, Stanford University. scale.stanford.edu
SCALE added 868 papers from October 2025 to June 2026; 201 were randomized or quasi-experimental, and 32 met its standard for strong causal evidence, 28 on student-facing AI and 4 on educator-facing AI. With the 20 from the March 2026 review, the total is 52. Five of the new student-facing studies were set in U.S. K-12, against none in March. The review states that "Using AI to offload academic tasks can reduce student learning" and that "Positive student experiences also do not necessarily translate into stronger learning outcomes."
Of the eleven K-12 studies the review lists, five report no overall gain or no difference. Oreopoulos and colleagues, 6,997 students in grades 6 to 8 across 20 U.S. schools: AI slowed practice without improving overall delayed learning. Sungu and colleagues, 193 teachers and 2,816 students in grades 5 to 12 in Türkiye: teacher AI access reduced student motivation and confidence, left average exam scores unchanged, and lowered scores for students of lower-performing teachers. Lim and colleagues, 874 U.S. teachers: more assignments completed, no significant value-added gain. Two German studies (391 and 241 students) found no significant differences against teacher feedback or standard ChatGPT.
Six report gains, including Chung and colleagues (770 Taiwanese high school students, 0.15 standard deviations) and the LearnLM Team at Google and Fab AI (1,423 students in grades 7 and 8 in Sierra Leone, 0.258 standard deviations). The authors caution that the findings "should not be treated as broadly generalizable" across U.S. K-12 settings.
Study context. This is a secondary synthesis that applies its own screen, and several of the studies it lists are working papers or developer reports. Counting studies isn't meta-analysis, and the review itself makes no pooled estimate.
Leadership implication. I'd give this review to your assessment director and instructional technology director with one assignment before the next renewal cycle: list each AI product you pay for and whether any study in the review, or any comparable trial, tested it in a setting like yours. Where the answer is no, require a pilot clause with a comparison group and a delayed learning measure. Brief your board that time saved and engagement aren't learning, using the review's own takeaway that perceived experience can rise while outcomes don't.
6. A Randomized Experiment at an English Sixth-Form College Found No Difference in Challenge-Seeking Across Human Tutoring, an AI Tutor, and Internet Search, With Human Tutoring Producing the Most Attempts
Source type. Peer-reviewed journal article; British Journal of Educational Technology, advance online publication November 13, 2026.
Citation (APA). Ennion, M., & McLellan, R. (2026). Human tutoring, generative AI tutoring, and independent internet search: A randomized between-subjects experiment on students' responses to academic challenge. British Journal of Educational Technology. Advance online publication. doi.org
Megan Ennion and Ros McLellan of the Faculty of Education, University of Cambridge, randomly assigned students aged 16 to 18 at an English sixth-form college to one of three supports while they worked on two sections of challenging mathematics: one-to-one human tutoring, a generative AI tutor, or unguided internet search. They measured effort, perceived difficulty, questions attempted, post-task stress, and later challenge-seeking.
By the second section, students with the AI tutor reported higher effort and greater perceived difficulty than students with a human tutor. Stressed responses were descriptively lower with the AI tutor than with internet search. Human tutoring produced the lowest perceived difficulty and the most questions attempted in both sections, and descriptively the fewest stressed responses. However, the abstract says significant pairwise differences varied by section.
Challenge-seeking didn't differ by condition and was linked mainly to students' existing dispositions. The authors conclude that the AI tutor offered support distinct from both human tutoring and search, with potential value during independent practice, and not as a replacement for human teaching. They argue that AI-supported learning should be judged by students' behavioral and affective responses to difficulty, not only by performance.
Study context. The abstract doesn't report the sample size, effect sizes, or p-values, and I couldn't open the article. It's a single college, a short task, and mostly self-reported measures, so it can't speak to achievement over a term.
Leadership implication. When a vendor proposes an AI tutor as a substitute for in-person intervention time, I'd have your curriculum director ask what the tool does to persistence, not just scores. Write a pilot measure into the agreement that counts questions attempted and students' willingness to take on harder work, and compare it with your existing human tutoring, not with no support. Keep human tutoring in the intervention budget until your own comparison says otherwise.
7. A Peer-Reviewed Analysis of Three Student Wellbeing Surveillance Tools Finds Vendor Claims of Safety With No Published Outcome Data, and Names Learning, Psychological, and Equity Harms
Source type. Peer-reviewed journal article, qualitative and exploratory; Postdigital Science and Education, November 6, 2026, open access.
Citation (APA). Nemorin, S. (2026). "Hey, I see you. Do you need help?" Understanding the algorithmic gaze of AI-driven student surveillance. Postdigital Science and Education. doi.org
Selena Nemorin of Monash University searched the EdSurge EdTech Index, an Australian education technology directory, and popular media from January to April 2026, identified 23 companies, and screened them to three tools with an Australian presence and a mental health and wellbeing focus: GoGuardian Beacon, Halo Education Analytics, and Saasyan Assure. She applied a thematic analysis to their marketing pages, customer story blogs, and support documentation.
The paper identifies three kinds of harm. Learning harms: engagement turns into compliance, higher-order thinking narrows, and students self-censor. Psychological harms: constant monitoring produces stress and performance for the algorithm, and Halo's tracking of faith formation turns spiritual life into a metric. Equity harms: biased training data can produce false positives and disproportionate flagging of Black students, LGBTQI+ students, students with disabilities, and students from low-income households who rely on school devices at home. The author argues the algorithmic gaze "does not simply predict student outcomes; it can actively shape them."
The paper doesn't report accuracy, false-positive rates, or student outcomes for any tool, because the materials it analyzed don't contain them.
Study context. The author states the limits: three tools, company-produced materials likely to be favorable, and no empirical work with students. It identifies potential harms; it can't show how often they occur in any school.
Leadership implication. I'd have your student services director and chief technology officer list every monitoring or alerting tool on student devices, including any bundled into a filter you already own. For each, request in writing the false-positive rate, the human review step before an alert reaches staff, the categories it flags, the retention period, and any opt-out. Where the vendor can't answer, report that to your board as an open risk, and set a review date before renewal.
8. Interviews With 29 Young People Who Use AI for Emotional Support Found It Doesn't Have to Be Trustworthy, Only Less Risky Than a Person, and That Most Turned to It After a Rupture in a Human Relationship
Source type. Nonprofit qualitative research report, not peer-reviewed, as reported by The 74 on November 10, 2026.
Citation (APA). Toppo, G. (2026, November 10). AI needn't be trustworthy, teens say. It just has to be less risky than people. The 74. the74million.org
The 74 reports a follow-up study by Alison Lee, chief research and development officer at the Rithm Project, and colleagues. A March survey of 2,383 young people ages 13 to 24 found 18 percent use AI for personal and emotional support and 15 percent talk to AI characters or personas. The researchers then analyzed 29 in-depth interviews with self-selected respondents from those two groups, ages 16 to 24.
Users didn't set out to confide in a chatbot. They started with homework, grammar checks, or casual conversation, and most shifted after what the report calls a gateway moment, usually a rupture in a human relationship. Young people understood sycophancy and hallucination and judged them manageable; human judgment felt riskier. The core finding, as The 74 quotes it: AI "doesn't have to be trustworthy. It just has to be less risky than a real person." Some hid their use from parents out of fear of losing access. Roughly one in four showed little or no awareness of AI's problematic behaviors, and participants reported little meaningful guidance from adults.
Study context. The researchers say the interviews are a self-selected set of case studies, not representative of students or children, and the sample includes young adults up to 24.
Leadership implication. I'd have your student services director add one question to your next student climate survey on whether students use AI for personal or emotional support, and report the result by grade band. Publish a plain-language route for students and families to reach a person when they're struggling, and review every product that claims to offer emotional support before it reaches a student device. Separate teacher-guided academic AI from companion-style tools in your policy, by name.
Emerging Strategic Themes
Theme 1. The state names the topic, and the board writes the clause. Kansas listed artificial intelligence as one of five required areas and published no sub-elements in its release. Edition 41 found state guidance copied across 28 states with no version history, and Edition 39 watched Maryland's deadline pass without a named enforcer. A topic on a state list protects no student until a local board turns it into an owner, a tool list, and an evidence test. Draft those three before your state tells you the date.
Theme 2. A union and vendor standard is a floor you still have to read. Edition 33 recorded the Microsoft memorandum and its exclusion of general-purpose tools. This week, the same standard reached Anthropic: Claude for Teachers only, with a research and safety exception to the training ban and no audit right named. Two vendors under one standard doesn't mean two identical contracts, and a tool built on a covered model isn't covered. Compare each agreement term by term with your own before telling your board a product is protected.
Theme 3. An off switch you can't flip separately is a single product. Edition 37 called the off switch a procurement specification, and Edition 41 recorded Google moving its default. This week Seattle blocked Search AI features, and Chicago said it would look elsewhere if it can't. When a basic utility and an AI feature ship together, a district is buying both whether it wants them or not. Keep a feature-level register and score separability in every request for proposals.
Theme 4. Feeling supported is not being safe or learning more. Stanford's review says positive student experiences don't necessarily mean stronger outcomes. Rithm Project interviews found young people use AI because it feels less risky than people, and a Monash analysis found wellbeing monitoring sold on care with no error rates. Edition 40 warned that self-report is not competence. Require a delayed learning measure in every pilot, an error rate from every monitoring tool, and a human route for every student who needs one.
What Was Not Found
The text of the Kansas policy and Board Directive No. 2026-1 wasn't located. A search of the KSDE release, news center, and board materials turned up neither the November policy nor the directive. The release lists five areas and no sub-elements, and plan details have appeared only in news coverage. Kansas districts are starting to draft against a deadline without the text that defines what the AI section must contain. New this week.
No peer-reviewed K-12 trial with a standardized achievement outcome carried a verified in-window date. The research sweep checked target journals by online date and arXiv's November listings. The nearest peer-reviewed designs report effort, stress, or self-reported dispositions. Zhou, Fu, Fang, and Hwang (British Journal of Educational Technology, online November 13) ran a three-week quasi-experiment with two intact grade 5 classes in eastern China and found a modest significant gain in self-reported computational thinking tendencies and no significant difference in algorithmic knowledge. Districts are adopting tutors at scale, but they rarely measure what students know a month later. Persistent gap.
No accuracy, false-positive, or outcome study of a K-12 wellbeing monitoring tool appeared in the window. I ran the counter-evidence sweep for student monitoring, engagement detection, emotion recognition, and proctoring. The only in-window study was the Monash documentary analysis, which reports that vendor materials contain no such figures. What a flag does to the flagged student remains unmeasured while these tools run on school devices. Persistent gap.
No in-window study measured AI effects for English learners or students with disabilities. Searches for algorithmic bias in automated scoring, AI tutoring outcomes by English learner status, and disability accommodations returned conceptual essays but no K-12 outcome studies from the window. The Monash paper names students with disabilities among those likely to be over-flagged, as a potential harm, not a measured one. Accommodation and monitoring decisions for these students are being made without outcome data, a persistent gap carried from Editions 40 and 41.
No audit, compliance report, or enforcement action under the union and vendor standard has been published. The Microsoft protections have been available to districts since November 1, and the Anthropic agreement took effect November 10. I found no published audit, certification, or complaint under either, and the Anthropic announcement doesn't describe an enforcement mechanism. Districts are being told a product is covered without a public record that coverage has been tested. New this week.
No Georgia district AI instrument could be verified inside the window. Cobb County's November 12 agenda carried no AI item; its policy items were JBC on school admissions and the 2027 legislative priorities.
The pattern is that obligations and standards are arriving ahead of their own texts and ahead of the evidence a board needs to meet them. A state deadline arrives as a release, a binding agreement arrives as a press page, and the strongest new review finds most recent K-12 trials flat. At the federal level, the sweep found no FTC, Department of Education, or GAO action on AI and schools in the window, and a search for district-wide deployments of autonomous AI agents returned vendor marketing pages and no evaluated deployment. That matters because Claude for Teachers includes agentic tools: an agent that can change records or send messages needs a named owner, limited permissions, logging, human approval for consequential actions, and a tested stop. The correct response isn't delay. It's monitoring, documentation, and exit ramps: a dated inventory of every AI tool, a written request for every text you're told binds you, and a pilot end date for each tool that can't show a comparison.
Novo Executive Summary
On November 13, Kansas announced that every district and accredited nonpublic school must adopt an instructional technology policy covering artificial intelligence by July 1, 2027, but has not yet published the policy text. On November 10, the AFT and UFT made their privacy standard binding on Anthropic, for Claude for Teachers only, with no audit right named. Seattle blocked Search AI features, Chicago said it may leave Google search, and California's superintendent and teachers' federation took opposite positions within two days. Stanford's review found that offloading academic work to AI can reduce learning, a Cambridge experiment found no difference in challenge-seeking, and analyses of monitoring tools and of young people's emotional use of AI found care claimed and not measured. None of this is solved by choosing the right product. It's solved by architecture: decision rights for each area Kansas named, a procurement specification that tests every standard and off switch, an evaluation plan with a delayed learning measure, and role-based literacy pathways. Novo Innovative Pathways works with district leaders to build that governance architecture, role-based AI literacy, and implementation strategy, so a board can adopt a policy it can defend clause by clause.
Watch This Week
- Wednesday, November 18 through Friday, December 4, 2026, U.S. Bankruptcy Court, District of Delaware, Case No. 26-11603 (BLS). Per PPC Land, on November 15, Smirks and Pangs Acquisition LLC was designated as an additional stalking horse for the non-Disney assets on November 14 at $10,000,000 in cash (Docket No. 164), and competing proposed orders on a consumer privacy ombudsman were filed on November 12 (Docket No. 148). Objections to that designation are due November 18 at 9:00 p.m. Eastern; bids are due November 24 at noon, the auction is December 1, and the sale hearing is December 3 and 4. Carried forward from Editions 38, 40, and 41.
- Thursday, November 19 and Monday, November 30, 2026, Google. New York City Council Speaker Julie Menin and Council Member Shekar Krishnan asked Google for written answers by November 19 on whether students can use Search without AI Mode, AI Overviews, or other generative AI tools; Senator Edward Markey asked by November 30 whether schools may disable AI Overviews, Gemini, and Guided Learning while keeping Search. A Google spokesman told reporters on November 4 that Gemini and Search are separate controls. Watch also for Chicago Public Schools' decision on alternative search tools. I found no published reply to the senators' October 21 letter. Carried forward from Editions 39 through 41.
- Los Angeles Unified Generative AI Ad Hoc Committee. LAist reported on November 9 that the committee would meet Friday, November 13, from 10 a.m. to 1 p.m., and that the district has asked Google to turn off AI-generated search summaries. I found no report of what the committee did. Correction: Edition 41 said a published schedule listed December 9 as the next meeting. Carried forward from Editions 32 through 41.
- Monday, November 16 through Monday, December 7, 2026, Georgia boards. Forsyth County holds its monthly regular meeting November 16; the Atlanta Public Schools Policy Committee meets November 18 and the board December 7, per the district calendar recorded in Edition 41; Gwinnett meets November 17 and 19 and Cherokee November 19. Whether Atlanta's draft instructional AI policy reaches a vote is unconfirmed, and no AI item is confirmed on any agenda. Carried forward from Editions 32 through 41.
- Wednesday, December 2 and Wednesday, December 16, 2026, Anne Arundel County Public Schools; and January 2027, Katy ISD. A district article dated October 16 lists Anne Arundel's second reading of the draft AI policy on December 2 and adoption on December 16. In Katy, the Houston Chronicle reported on November 12 that a community committee proposed AI literacy, screen-time, online safety, and AI efficiency objectives among 21 new objectives; trustees didn't vote, and budget development begins in January. Carried forward from Editions 35 through 41; Katy new.
- Before Friday, January 1, 2027, Colorado Attorney General and New York; and Thursday, July 1, 2027, Kansas. Colorado must adopt the rules for SB 26-189 and HB 26-1263 before the statutes take effect January 1, and the Attorney General's page showed no final adoption. New York's S9051-B, Chapter 368, takes effect January 1. Kansas district policies and plans are due July 1, 2027. In California, the CFT's request for a three-year K-8 pause awaits the outgoing governor, who hadn't responded as of November 11. Carried forward from Editions 38 through 41.
- Paris Peace Forum follow-through. The Forum's November 10 release lists 57 signatories to the Children First Declaration, with no United States signature, and 8 million euros pledged by OpenAI, Anthropic, and Common Sense Media for AI and child development research. It is voluntary and creates no obligation for a U.S. district. Carried forward from Editions 35 and 41.
Questions District Leaders Are Asking
What does Kansas now require school districts to do about AI?
A Kansas State Department of Education release dated November 13, 2026 says the State Board approved a policy, a resolution, and Board Directive No. 2026-1. Every district and accredited nonpublic school must address five areas in an instructional technology policy by July 1, 2027: purpose and intended use of technology, student access and use, digital citizenship and student wellbeing, artificial intelligence, and privacy and security. The release lists no sub-elements for artificial intelligence, and the policy text and the directive were not located.
Does the Kansas policy require parent consent for school technology?
The release describes acknowledgment, not consent. Districts must establish a process for parents to acknowledge the school's policy by the start of the 2028-29 school year, and the release does not mention consent or an opt-out. The Johnson County Post reported on November 9 that the board was weighing acknowledgment against active consent and that five Johnson County districts opposed consent in an October 16 joint letter.
What does the AFT and UFT agreement with Anthropic cover?
The AFT and the United Federation of Teachers announced a legally binding agreement that applies the national AI Safety and Privacy Standard for Schools to Anthropic's existing district agreements for Claude for Teachers. Education Week reports that it bars the use of student, educator, and customer data to train or improve models, with exceptions for certain research and safety-related data. It covers only Claude for Teachers, not other companies' products built on Claude models, and neither source describes an audit right, an enforcement mechanism, or a penalty.
Can a school district turn off the AI features inside Google Search?
That is the open question. KOMO reports that Seattle Public Schools began blocking Google Search AI features, Gemini, and unapproved generative AI tools at every grade on November 9, and no source explains how Seattle does it. A Google spokesman told reporters on November 4 that Gemini and Search are separate controls, both off by default for K-12 student accounts. Senator Markey asked Google to answer by November 30 whether schools may disable AI Overviews, Gemini, and Guided Learning while keeping Search, and the New York City Council asked by November 19.
What did Stanford's November 2026 evidence review find about AI in K-12?
The SCALE Initiative review, published November 10, 2026 and not peer-reviewed, added 32 studies with strong causal evidence, 28 on student-facing AI and 4 on educator-facing AI, for 52 in total. Five of the new student-facing studies were set in U.S. K-12. The review states that using AI to offload academic tasks can reduce student learning and that positive student experiences do not necessarily translate into stronger learning outcomes. Of the eleven K-12 studies it lists, five report no overall gain or no difference.
What should a district outside Kansas do now?
Read Kansas as a template. Name an owner for each of the five areas now, keep a dated inventory of every AI tool in use, and write the evidence question into the policy, not the handbook. In the AI section, list every student-facing AI tool, name who approves each one, and state what evidence it must show at renewal. A district that waits for its state to publish its own version will write the policy under a deadline someone else chose.
Sources
Governance and Policy
Kansas State Department of Education. (2026, November 13). Kansas State Board of Education requires all school districts to develop instructional technology policies; policy prioritizes high-quality instruction while preserving local decision-making [News release]. ksde.gov
Gregory, D. (2026, November 13). Kansas schools must develop new instructional technology policies by 2027. KSNT, via Yahoo News. yahoo.com
Mellott, M. (2026, November 9). JoCo school districts oppose state tech opt-out proposal. Johnson County Post. johnsoncountypost.com
American Federation of Teachers. (2026, November 10). AFT, UFT and Anthropic sign binding AI safety and privacy agreement for schools [Press release]. aft.org
Sparks, S. D. (2026, November 10). Anthropic agrees to protections for teachers who use Claude. Education Week. edweek.org
Modan, N. (2026, November 10). Anthropic enters student AI privacy agreement with teachers unions. K-12 Dive. k12dive.com
KOMO News Staff. (2026, November 5). Seattle schools to cap elementary students' screen time, block unapproved AI tools. KOMO News. komonews.com
Gasca, N. (2026, November 5). Seattle Public Schools tightens student access to YouTube, Minecraft and AI. KUOW. kuow.org
Seminera, M. (2026, November 12). Protest outside of Chicago Google office points to rising concerns about AI in schools. Chalkbeat Chicago. chalkbeat.org
Baron, E. (2026, November 4). Google reverses decision to turn off search for schools not using AI. Tribune Content Agency, via Government Technology. govtech.com
California Department of Education. (2026, November 9). State Superintendent Tony Thurmond convenes summit on AI in education, emphasizes innovation, student safety, and educator leadership (Release #26-46). cde.ca.gov
Barth, M. (2026, November 11). California teachers union presses Newsom for three-year AI classroom pause. California Globe. californiaglobe.com
Goodman, C. (2026, November 12). Katy ISD committee proposes AI lessons, screen-time limits and stronger online safety. Houston Chronicle, via Yahoo News. yahoo.com
Office of Senator Edward J. Markey. (2026, November 9). Markey demands Google allow schools to disable AI services without blocking search access [Press release]. markey.senate.gov
New York City Council. (2026, November 13). Press release on Speaker Menin and Council Member Krishnan letter to Google. council.nyc.gov
Yu, E. (2026, November 9). LAist report on Los Angeles Unified and Google AI search. LAist. laist.com
PPC Land. (2026, November 15). Character.AI users' chats face a $10M sale to Smirks and Pangs. ppc.land
Paris Peace Forum. (2026, November 10). A global commitment to secure a better future for every child [Press release]. parispeaceforum.org
Georgia, Maryland, and Colorado records: Cobb County School District board agenda, November 12, 2026, portal.cobbk12.org; Forsyth County Schools meeting schedule, forsyth.k12.ga.us; Anne Arundel County Public Schools article, October 16, 2026, aacps.org; Colorado Department of Law, coag.gov; Bianchi, R. (2026, September 17), Atlanta News First, atlantanewsfirst.com
Research, Peer-Reviewed
Ennion, M., & McLellan, R. (2026). Human tutoring, generative AI tutoring and independent internet search: A randomized between-subjects experiment on students' responses to academic challenge. British Journal of Educational Technology. Advance online publication. doi.org
Nemorin, S. (2026). "Hey, I see you. Do you need help?": Understanding the algorithmic gaze of AI-driven student surveillance. Postdigital Science and Education. doi.org
Zhou, W.-L., Fu, Q.-K., Fang, J.-W., & Hwang, G.-J. (2026). Generative AI empowered contextual educational games as facilitators of elementary school students' computational thinking tendencies. British Journal of Educational Technology. Advance online publication. doi.org
Research, Preprint (Not Peer-Reviewed)
None this week. The research sweep surfaced arXiv preprints dated in the window. Still, none had a K-12 sample with a measured student or teacher outcome that I could confirm: one reported withdrawn classroom outcome tables, and others had no K-12 sample or no stated sample. None is cited.
Institutional Report (Not Peer-Reviewed)
Walker, S., Fesler, L., Martinez Claeys, J. P., Agnew, C., & Loeb, S. (2026, November 10). The evidence base on AI in K-12: A November 2026 review [Institutional evidence review, not peer-reviewed]. SCALE Initiative, Stanford University. scale.stanford.edu
Toppo, G. (2026, November 10). AI needn't be trustworthy, teens say. It just has to be less risky than people. The 74. the74million.org
Kansas has set a date before publishing the text, and a binding vendor standard arrived as a press page. Neither waits for your next board cycle. The Novo 10-Domain Readiness Brief is where a district starts: a named owner for each policy area, a dated inventory of every AI tool, a feature-level register of what can be switched off, and a pilot clause with a delayed learning measure. Get The Novo 10-Domain Readiness Brief at no cost, or schedule a conversation. We build that architecture with district leaders, domain by domain.
Schedule a Readiness Conversation