California Releases a Model AI Policy as Ohio's First Hard Deadline Arrives: The Template and the Timer

Framing

The governance story this brief has tracked all spring has been a climb. Maryland assigned ownership. Connecticut required instruction. The Senate, two weeks ago, offered funding and a study, but no federal standard. This week, the action returns to the states, and it arrives in the two blunt instruments that actually move district behavior: a model policy and a deadline. On June 25, the California Department of Education published the Model Policy on Artificial Intelligence in Education required by Senate Bill 1288. Three days later, on July 1, Ohio's statutory requirement that every district adopt an AI policy takes effect. The largest state system in the country issued a template in the same week that the first hard compliance clock in the country runs out.

Read together, California and Ohio define the two ends of state influence. A model policy is an offer. A deadline is an order. Both are now in play, and both, notably, center the same operational core: procurement of tools that protect student and educator data, and use that supports rather than substitutes for teaching. Neither instrument, by itself, builds the thing a district actually needs. A model policy is a floor that a district can adopt without changing a single practice. A deadline produces a document by a date, not a capability by design. The distance between having a policy and governing AI is exactly the distance this brief exists to name.

The counter-pressure sharpened in the same window. Twenty-nine members of the New York City Council wrote on June 9, demanding a two-year moratorium on AI in the nation's largest district, arguing that the city's draft guidance does not adequately protect student data, and noting that a national coalition has called for a five-year pause on student-facing generative AI. The research that surfaced this week says the evidence cannot yet adjudicate any of it. An international audit found that commercial language models do not faithfully represent teachers' own views of AI's benefits and risks. The newest classroom design paper proposes a multi-year trial that has not been run. A peer-reviewed expert panel could agree on only a third of what AI literacy means. States are standardizing, and activists are demanding pauses, on top of an evidence base that is still being designed. The districts that come through this well are the ones that treat a model policy and a deadline as the beginning of governance architecture, not the end of a compliance task.

Top Research and Policy Signals

1. California Releases Its SB 1288 Model AI Policy, Putting a State Template in Front of the Nation's Largest School System

Source type. State agency guidance (model policy; California Department of Education; released June 25, 2026; non-binding).

California Department of Education. (2026). Model policy: Artificial intelligence in education (Senate Bill 1288, 2024). cde.ca.gov

On June 25, 2026, the California Department of Education released the model AI policy required by Senate Bill 1288 (2024), developed by the state's AI in Education Workgroup, a group of California teachers, educators, and subject-matter experts, with public comment. The document is designed to help districts use AI in ways that benefit and do not harm students and educators, and it foregrounds the procurement of software that protects the safety and privacy of students, educators, and their data, alongside effective use that supports teaching and learning and avoids risk.

Critically, it is a model, not a mandate. California districts retain local control over whether and how to adopt it. With roughly 5.8 million public school students, California's template will shape vendor behavior and neighboring-state expectations well beyond its borders.

Leadership implication. A state model policy is the most useful and the most dangerous document a cabinet will see this year. Useful, because it is credible scaffolding. Dangerous, because it is adoptable without changing anything. Treat the California model as a checklist of what to localize, especially procurement and data-privacy criteria, not as a policy to ratify and shelve. The districts that win convert the template into binding purchase gates and named accountability; the rest will have a compliant document and an ungoverned practice.

2. Ohio's July 1 Deadline Arrives: The First Hard State Clock on District AI Policy Runs Out

Source type. State law and agency model policy (Ohio House Bill 96; Ohio Department of Education and Workforce model policy, released January 2026).

Ohio Department of Education and Workforce. (2026). AI model policy for Ohio districts and schools. education.ohio.gov  See also Kohrman Jackson & Krantz. (2026, June 12). Ohio's July 1, 2026, school AI policy deadline. kjk.com

Under House Bill 96, every traditional public school district, community school, and STEM school in Ohio must adopt a formal AI policy by July 1, 2026. The state's Department of Education and Workforce released a model policy in January 2026 that districts may adopt or customize. It emphasizes AI literacy; structured governance through an ongoing AI workgroup representative of grade levels and special education; family and community engagement; and data privacy compliance covering personally identifiable information, FERPA, and applicable state and federal law.

Ohio is among the first states in which the requirement is not a recommendation or a future guideline, but a dated obligation that takes effect this week.

Leadership implication. A deadline manufactures documents, not governance, and the July 1 cohort proves the point at scale. The question for any Ohio cabinet is not whether a policy exists by Tuesday but whether the policy names who owns AI decisions, what a tool must prove before purchase, and how the workgroup will keep the policy current as tools change monthly. For leaders in the other 49 states, Ohio is a preview: assume your deadline is coming and build the architecture now, while you can do it deliberately rather than under a clock.

3. An International Audit Finds Commercial AI Models Do Not Mirror Teachers' Own Views of AI

Source type. Preprint (not peer-reviewed); arXiv; presented at the ACM Conference on Learning @ Scale, June 2026. Analysis of OECD TALIS teacher-survey data across 55 countries and territories with systematic model evaluation.

Authors as listed on arXiv (names not confirmed from full text; flagged). (2026). Teachers' perceived benefits and risks of AI across fifty-five countries: An audit of LLM alignment and steerability [Preprint]. arXiv:2605.08486. arxiv.org

The study pairs representative international teacher-survey data (OECD TALIS, 55 countries and territories) with a systematic evaluation of whether large language models reproduce teachers' perceived benefits and risks of AI. The models did not align well with teachers' actual views. Prompting a model with country-specific context narrowed the gap between model and human responses. Still, it did not close it, and only some models approximated the relative ranking of teachers' values.

The authors note that large language models are already used in research, policy, and teachers' workflows despite limited validation in education settings.

Leadership implication. If the model your district is piloting does not reliably represent what your teachers consider beneficial or risky, then asking the AI is not a substitute for asking your educators, and a vendor's claim of pedagogical alignment is a hypothesis, not a fact. Build educator judgment into procurement and into any AI-assisted policy drafting, and require evidence of validation in education settings, not general benchmarks. This is a preprint that has not yet cleared peer review; weigh it as a serious signal, not a settled finding.

4. The Newest Classroom Design Work Proposes a Multi-Year Trial That Has Not Been Run

Source type. Preprint (not peer-reviewed); arXiv; June 2026; system design and proposed longitudinal evaluation (Northern Arizona University).

Authors as listed on arXiv (names not confirmed from full text; flagged). (2026). AI-integrated learning management system for middle school: A longitudinal study of learning outcomes through high school and beyond [Preprint]. arXiv:2606.07544. arxiv.org

The paper proposes an AI-integrated middle-school learning platform built around bounded, policy-gated assistance that gives formative feedback and hints rather than answers, mastery-based and spaced practice, teacher dashboards that flag sustained struggle, and a privacy-first design with data minimization, role-based access, and audit logs.

Just as important is what it does not contain: outcomes. The authors lay out a longitudinal evaluation framework, preferably a cluster-randomized trial or stepped-wedge rollout, to test whether early AI support produces durable gains through high school and beyond, explicitly because the field lacks evidence past short-term engagement, and fade-out is a known risk. This is a blueprint for measuring impact, not a report of impact.

Leadership implication. The most recent academic work in this space still proposes ways to determine whether AI helps over time, which should inform every adoption decision a cabinet makes this summer. The design principles are worth borrowing now: bounded assistance, teacher-in-the-loop dashboards, audit logs, and equity analysis by subgroup, because they are governance requirements you can specify in a contract regardless of outcomes. But treat any vendor promise of durable learning gains as unproven, because even the researchers proposing to measure it concede the evidence does not yet exist.

5. A Peer-Reviewed Expert Panel Cannot Yet Agree on What AI Literacy Means, Just as States Write It Into Law

Source type. Peer-reviewed; Interactive Learning Environments (published online April 1, 2026). International Delphi study; expert panel across three rounds (n = 33, 27, 25). Dated earlier than this week's window and included as the most recent verifiable peer-reviewed anchor.

Authors as listed (names not confirmed from full text; flagged). (2026). AI literacy for K-12 education: An international Delphi study. Interactive Learning Environments. doi.org

An international, interdisciplinary expert panel ran three Delphi rounds to define the central components of AI literacy in K-12 education. After three rounds, the panel reached consensus on only 8 of 23 candidate statements, organized under two themes: foundational AI knowledge and critical perspectives.

The study is among the most recent peer-reviewed attempts to pin down a construct that, in the authors' framing, remains loosely defined even as it becomes a policy priority worldwide.

Leadership implication. California, Connecticut, and Maryland are now mandating AI literacy in standards and instruction, while the research community agrees on only a fraction of what the term contains. That gap is a curriculum risk: a district that teaches AI literacy without specifying which components will produce inconsistent, unassessable instruction. Define your own working construct, anchored in foundational knowledge and critical evaluation, and make it concrete enough to teach and measure, rather than waiting for a consensus that does not yet exist.

Emerging Strategic Themes

Theme 1. Templates and Timers. State influence over district AI now travels through two instruments: model policies (California) and deadlines (Ohio). Both are arriving at once, and both are easy to satisfy on paper. The strategic risk is compliance theater, a ratified template, or a met deadline that leaves actual practice ungoverned. Plan for the instrument your state will use, and build past it.

Theme 2. The Procurement Chokepoint. Every serious document this week, California's model, Ohio's model, and the middle-school design paper, converges on procurement and data privacy as the operative control. This is the most leveraged governance surface a district has: what you refuse to buy, you never have to govern. Districts that locate their AI governance at the point of purchase will outperform those that try to police use after deployment.

Theme 3. The Moratorium Counter-Current. Alongside mandates, organized demands to pause are intensifying, with 29 New York City Council members for two years and a national coalition for five. Whether or not any pause is adopted, the political reality is that adoption now generates organized resistance. Cabinets need a defensible, evidence-based position on pace, not just on tools, because asking why we are moving this fast is now a board meeting question.

Theme 4. Tools That Do Not Speak for Teachers. The 55-country audit is a warning about a quiet substitution, using AI to stand in for educator judgment in research, policy drafting, and classroom decisions. Models do not reliably reflect what teachers value. Keep human educators in the loop on the decisions that shape policy and practice, and treat the AI as input, never authority.

What Was Not Found

This week produced two governance instruments and three research artifacts, and not one of them is causal outcome evidence. No peer-reviewed study within the seven-day window reported whether any AI tool improves achievement, narrows or widens gaps, or affects student well-being in U.S. K-12 settings. The most recent classroom design paper explicitly proposes a trial that would generate such evidence and concedes that it has not been run. The most recent peer-reviewed contribution we could verify is a consensus-definition exercise, not an outcome study.

The specific absences matter because districts are acting now. There is still no verified peer-reviewed evidence on whether a state model policy, like California's, changes what districts actually buy or how teachers actually use AI. There is no causal evidence that meeting a deadline, such as Ohio's, produces safer or more effective AI use than filing a document. There are no peer-reviewed outcome data this week for the populations at highest risk: English learners, students with disabilities, elementary readers, and students in high-poverty districts. And there is no evidence that the AI literacy now being written into law in California, Connecticut, and Maryland improves any student outcome, in part because, as this week's Delphi study shows, the field cannot yet agree on what AI literacy is. Adoption and mandate are both running well ahead of evidence. That is not a reason to stop. It is the reason to govern.

Novo Executive Summary

California handed the nation's largest school system a model AI policy this week, and Ohio's first hard deadline arrives three days later, which means the two dominant levers of state influence, the template and the timer, are now both in motion. But a model policy is a floor, and a deadline is a due date, and neither is governance. The evidence base, as this week again confirms, is still being designed rather than reported, while organized voices demand the brakes be applied. In that environment, the durable advantage belongs to districts that convert a template and a deadline into architecture: named ownership, procurement gates that protect student data, role-based AI literacy defined concretely enough to teach, and outcome measures chosen before tools are bought. That is precisely the work Dr. Reginald Griffin and Novo Innovative Pathways do with district cabinets, turning compliance obligations into governance capability. We don't sell AI. We govern it.

Watch This Week

• Ohio's July 1, 2026, deadline: every district, community, and STEM school must have an adopted AI policy. Watch compliance rates and the quality gap between adopted templates and localized policies.
• California districts begin working from the SB 1288 model policy. Watch whether county offices and large districts localize procurement and data privacy criteria or adopt them verbatim.
• New York City's comprehensive AI Playbook remains pending after the March preliminary guidance. Watch its release against the 29-member City Council's push for a two-year moratorium.
• The national coalition calls for a five-year pause on student-facing generative AI. Watch whether any large district formally entertains a moratorium resolution.
• Peer-reviewed outcome studies: watch Computers & Education, the Journal of Learning Analytics, and npj Science of Learning for causal K-12 results to replace this week's preprints and proposals.
• Summer procurement season: watch for districts signing AI contracts ahead of fall without the data-privacy and validation criteria that this week's sources all flag.

Sources

Governance and Policy

Research, Peer-Reviewed

Research, Preprint (Not Peer-Reviewed)

AI in Public Education Brief is published weekly by Novo Innovative Pathways. For district advisory engagements, contact Dr. Reginald Griffin through Novo Innovative Pathways.