Students From All 50 States Pass a Model AI Act for Their Own Schools: The Constituency Map of K-12 AI Governance Is Complete

Framing

For two years, the question hovering over every district AI policy has been some version of: who gets a say? In 2026, the answer has arrived in layers. Legislatures set statutory floors. State agencies wrote model policies and rules. School boards adopted handbooks. In July, organized labor claimed its seat through the AFT's convention resolution. Last week, Florida showed that compliance documents districts already own can carry AI governance. This week the final missing constituency stepped forward: the students themselves.

On August 3, AASA announced that students from all 50 states, convened as a National AI Student Senate at the Edward M. Kennedy Institute's full-scale replica of the United States Senate chamber, had drafted, debated, and passed the STUDENTS FIRST Act of 2026 by a vote of 82 to 16. The effort was organized by Day of AI, MIT RAISE, AASA, and the Kennedy Institute. This is not a petition. It is a structured model act, and its provisions converge, without coordination, on the same architecture questions running through statehouses: due process when students are accused of improper AI use, human review of consequential decisions, prohibitions on profiling, disclosure of data collection, and the right to decline AI with a real alternative. When the governed independently arrive at the same governance questions as the governors, those questions stop being optional agenda items. They are the invariant structure of K-12 AI governance.

The week's research supplies the uncomfortable counterpoint: the capacity to answer those questions is thin. Peer-reviewed findings this week show adaptive tutoring licenses producing no class-level gains where implementation was left to chance, district policy development proceeding by improvisation, teacher trust in monitoring tools depending on design choices few procurement offices know to specify, and teacher readiness so segmented that one-size-fits-all training misallocates money by design. Students have now written down what they expect institutions to guarantee. The institutions, on this week's evidence, have not yet built the machinery to guarantee it.

The legislative calendar makes this the right moment to close that distance. This was the quietest legislative week of 2026, with California's roughly 30 surviving AI bills queued for twin suspense votes on August 13 and most other statehouses in recess. Districts that use the quiet to stand up governance architecture will meet the autumn wave of statutes, rules, and constituency demands on their own terms. Districts that wait will meet it on someone else's.

Top Research and Policy Signals

1. Students From All 50 States Pass a Model AI Act for Their Own Schools

Source type. Institutional policy framework and press release (not peer-reviewed). Announced August 3, 2026.

AASA, The School Superintendents Association. (2026, August 3). Students from all 50 states produce national framework for AI in America's schools. aasa.org
Framework text: Day of AI. (2026). The STUDENTS FIRST Act of 2026. dayofai.org

At America's Youth AI Festival (July 17 to 19, UMass Boston and MIT), select students formed a National AI Student Senate and passed a comprehensive model act, 82 to 16. The act would require AI literacy instruction before students first use school devices; prohibit AI-generated written and artistic assignments at every grade; permit supervised, disclosed AI use for brainstorming, studying, and editing from 9th grade with teacher permission; guarantee a right to appeal AI-misuse accusations with meaningful human review, so no grade or discipline rests solely on detection software; bar schools from using AI to profile students or independently determine grades, discipline, or hiring; require transparency about data collection; permit students to decline AI use and receive an alternative assignment; and draw a firm line against AI substituting for counselors or trusted adults. It rejects both an outright ban and unrestricted adoption. AASA will circulate the framework to its network of more than 10,000 school leaders, with webinars, workshops, and school visits planned.

Leadership implication. Treat this as a demand-side governance signal. When your board asks what students think, this document is now the default answer, whether or not your district has gathered its own. Superintendents should expect its due-process, opt-out, and anti-profiling provisions to surface in handbook debates this fall, and should decide deliberately which provisions to adopt, adapt, or decline, rather than encountering them for the first time in public comment.

2. Access Is Not Implementation: Two German Longitudinal Studies Find No Automatic Gains From Adaptive Tutoring

Source type. Peer-reviewed. Frontiers in Education, Original Research, published July 7, 2026.

Schaaf, J., Rolfes, T., Nagy, G., & Heinze, A. (2026). The effect of the frequency of use of an intelligent tutoring system on learning gains in mathematics in schools in challenging social circumstances. Frontiers in Education, 11. doi.org/10.3389/feduc.2026.1842708
Companion study: Schaaf, J., et al. (2026). The effect of the frequency of use of an intelligent tutoring system on learning gains in mathematics secondary education. Frontiers in Education. frontiersin.org

In 60 grade 8 and 9 mathematics classes serving schools in challenging social circumstances in Germany (587 students), researchers tracked a full school year of achievement tests, questionnaires, and system log data. Implementation was uneven and delayed: only 41 classes used the intelligent tutoring system at least once, and average use among users was low. Multilevel models found no class-level effect of usage on learning gains, and the average learning gain across the sample was statistically indistinguishable from zero. Students who used the system more than classmates during school time showed small positive associations with post-test performance; home use showed none. A companion study of 940 students in 55 classes in northern Germany found no significant effect of usage frequency at either the class or individual level. Usage itself was predicted by student work ethic, gender, students' rating of the tool, and teacher experience.

Leadership implication. These are the conditions most district adaptive-learning purchases quietly assume away. A license does not produce dosage, and dosage during school time is the only dosage that correlated with anything. Procurement should buy implementation conditions, not seats: embedded class time, teacher routines, and usage instrumentation with review checkpoints. In under-resourced settings, where these studies were run, the tool alone moved nothing.

3. Teachers Trust Monitoring Dashboards When the Dashboards Explain Themselves

Source type. Peer-reviewed conference proceedings. LAK26, 16th International Learning Analytics and Knowledge Conference.

Ooge, J., Faik, A., & Verbert, K. (2026). Detect, explain, act: How teachers trust and use an explainable real-time monitoring dashboard to detect student outliers in class. In Proceedings of LAK26: The 16th International Learning Analytics and Knowledge Conference. ACM. doi.org/10.1145/3785022.3785087

Researchers iteratively designed a real-time dashboard that flags outlier students on an adaptive learning platform and, critically, explains its detection logic in two registers: model-centric and data-centric. In a counterbalanced within-group experiment with follow-up interviews, 11 teachers used the dashboard in live classrooms. Teachers integrated it into instruction, and their trust was shaped by dispositional, situational, and learned factors. The data-centric explanations did the operational work: they let teachers validate whether a flag was accurate, check it against what they already knew about the student, and select an appropriate intervention rather than deferring to the system.

Leadership implication. Explainability has crossed from ethics language into procurement language. Districts can now point to peer-reviewed design evidence when RFPs require monitoring tools to expose the data behind every flag, because explanation is what converts an alert from an instruction into information a professional can verify. Note the scale honestly: 11 teachers is design-stage evidence, a reason for pilot clauses, not a warrant for district-wide claims.

4. Inside Six Districts: AI Policy by Improvisation

Source type. Peer-reviewed. Educational Policy, advance online publication.

McGuire, P., et al. (2026). Approaches to artificial intelligence policy development in K-12 school districts. Educational Policy. Advance online publication. [URL flagged for verification: publisher page not directly retrievable this week; record sourced via Consensus at consensus.app and full author list pending confirmation]

Interviews with leaders across six Colorado districts show how AI governance actually begins: informal guidelines, student-facing bans, or modifications bolted onto existing acceptable-use policies. The study documents persistent gaps in professional development and in structures for evaluating whether any of it works, and offers recommendations for leaders navigating the early stages of policy development. This is the third consecutive week Educational Policy has published district-level AI governance research, following the 424-district three-state analysis and the twelve-largest-districts typology featured in prior editions. The journal pipeline itself is a signal: district AI governance has become a named research field, and its first finding is that improvisation is the default.

Leadership implication. An acceptable-use paragraph is not an AI governance architecture. The research base now names the default failure mode precisely: rules without professional development and without evaluation attached. Cabinet teams should be able to answer two questions the literature keeps asking: who is trained to carry this policy into classrooms, and what evidence would tell us it is working.

5. Teacher Readiness Comes in Profiles, Not Averages

Source type. Peer-reviewed, open access. Universal Access in the Information Society, published online March 2, 2026.

Pallejà, I., Aguayo-Mauri, S., Fonseca, D., Iglesias, A., & Canaleta, X. (2026). Technological profiles of primary and secondary school teachers: A data-driven approach to AI training design. Universal Access in the Information Society, 25, Article 46. link.springer.com

Surveying 262 primary and secondary teachers in Spain during the 2024-25 school year, researchers combined correlation analysis with K-Means clustering across 71 variables and mapped the results onto the European DigCompEdu competence framework. Two macro groups emerged, active adopters and cautious or emerging adopters, which were refined into five micro profiles: advanced integrators, content and resource builders, data-and-code oriented users, safety-first fundamentalists, and skeptics or context-limited users. The authors' conclusion is architectural: readiness is segmented, so effective AI training must be targeted by profile rather than delivered uniformly.

Leadership implication. A single all-staff AI workshop misallocates professional development budget by design, because it trains five different populations as if they were one. The evidence-aligned sequence starts with a readiness diagnostic and routes staff into differentiated, role-based literacy pathways. Spanish context, but the segmentation logic transfers: your staff has these profiles whether or not you have measured them.

Emerging Strategic Themes

Theme 1. The governed are now governing. With the student framework joining statutes, agency rules, board policies, labor resolutions, and compliance documents, the constituency map around district AI decisions is complete. Districts are no longer writing policy into a vacuum; they are reconciling documented positions. Expect structured student input to become a legitimacy requirement for AI policy adoption, the way community input became one for strategic plans.

Theme 2. Implementation is the new effect size. The German tutoring studies join a season of findings in which outcomes track embedding, not licensing. The binding constraint on AI value in schools is instructional integration capacity: scheduled time, teacher routines, and usage review. Budget accordingly, and treat vendor efficacy claims that assume ideal dosage as claims about a district that does not exist.

Theme 3. Explainability is becoming a procurement specification. When peer-reviewed evidence shows explanations changing how teachers verify and act on system flags, requiring explanation interfaces stops being aspirational ethics and becomes contract language. Districts hold more drafting power here than they typically use.

Theme 4. The surveillance pipeline is outrunning the evidence pipeline. A single week's scan surfaced at least five new engineering-venue systems for real-time camera-based monitoring of student engagement or emotion, built on object detection, facial landmarks, and IoT sensors. They report detection accuracy; none reports learning outcomes, and privacy treatment is typically a design claim rather than a governance framework. These products will reach procurement offices within a resale cycle, and they collide directly with the profiling prohibitions students just drafted and Florida's proposed exclusion of such tools from instructional-tool definitions.

What Was Not Found

This section exists because adoption keeps outrunning evidence, and this week the distance was measurable in specific places.

• No causal evidence on student-voice provisions. The STUDENTS FIRST Act proposes appeal rights, opt-outs with alternatives, and disclosure duties. No study anywhere estimates whether such provisions change integrity rates, student trust, or learning. Districts adopting them this fall will be acting on legitimacy grounds, which is defensible, but they should not claim an outcome evidence base that does not exist.
• No false-accusation base rate. The single most consequential number in the AI-integrity debate is how many students per semester are wrongly flagged by detection tools in real districts. After two full school years of detector deployment, no peer-reviewed study reports it. Every appeal-rights provision now being drafted is calibrated blind.
• No outcome evidence behind classroom surveillance products. None of the camera-based engagement and emotion monitoring systems surfaced this week reports effects on learning or well-being, and none examines misclassification risk for English language learners or students with disabilities, populations for whom expression-based inference remains essentially unexamined.
• No federal evaluation instrument. The GAO study of AI in K-12 requested by Senate letter in June has not appeared, and no federal outcome evaluation of AI instructional tools is underway in public view. The federal posture remains permission and study, with the study still pending.
• Persistent gaps carried forward. Consent-default designs (Florida's proposed opt-in versus Oklahoma's opt-out) still have zero comparative evidence ten months into live statutory experiments. Elementary grades and literacy remain thinly studied relative to secondary mathematics, where this week's strongest implementation data again originated. And the conditions under which tutoring-system use becomes effective in high-poverty settings, the exact question the German null results pose, remain unanswered in any U.S. Title I context.

Novo Executive Summary

This week completed the constituency map: students joined legislatures, agencies, boards, and labor as authors of documented AI governance positions, and their model act converges on the same structural questions districts have been deferring. The research delivered the capacity audit: tutoring licenses without implementation conditions produced nothing; district policy is being improvised without training or evaluation; teacher trust depends on explainability, which few contracts specify; and staff readiness is segmented five ways. The strategic conclusion is that the differentiator in K-12 AI is no longer which tools a district selects but whether it owns the architecture that makes any tool governable: decision rights, procurement specifications, evaluation checkpoints, and literacy pathways matched to actual roles and readiness. That architecture is buildable now, during the quietest weeks the 2026 policy calendar will offer. Novo Innovative Pathways works alongside district leaders to build exactly this: governance architecture, role-based AI literacy, and implementation strategy that stand up to boards, bargaining units, and now the students themselves.

Watch This Week

• Thursday, August 13: California's Assembly and Senate Appropriations Committees hold twin suspense votes deciding the fate of roughly 30 AI bills, including AB 1159 (extending student privacy law to operators marketing to schools), AB 2656 (45-day union notice before public employers deploy generative AI in represented work), SB 867 (companion chatbots in toys), and the AB 2023 / SB 1119 chatbot-safety pair. Bills that fail cannot be revived this session.
• Michigan legislators return to Lansing this week, with kids chatbot safety bill SB 760 already through the Senate and awaiting House committee action.
• Florida: watch for the formal rulemaking notice on the proposed AI amendment to Internet Safety Policy Rule 6A-1.0957 following the August 5 development workshop, and for whether industry pressure narrows the definitions before districts inherit the January 1, 2027 compliance date.
• North Carolina: H 301, requiring a state model AI policy, a generative-AI tool evaluation framework, and educator training, sits in conference committee to reconcile House and Senate versions.
• New York: kids chatbot safety bill S 9051 and the AI Training Data Transparency Act A 6578 await Governor Hochul's signature, with a December 31 deadline.
• October: the teacher-student trust and agency study flagged in our July 26 watch list (Nagashima et al., arXiv:2607.01506) publishes in Proceedings of the ACM on Human-Computer Interaction (CSCW 2026), the first major venue treatment of whether teachers and students want the same things from classroom AI control.

Sources

Governance and Policy

Research, Peer-Reviewed

Research, Preprint (Not Peer-Reviewed)

No preprint is featured as a research signal this week. One preprint is watch-listed for its October peer-reviewed publication: Nagashima, T., Siegrist, L., Scholz, N., Sato, S., Vincoli, M., & Su, M. (2026). Mind the trust gap: Identifying (mis)alignments in teacher-student views toward control and agency in K-12 classroom AI. arXiv. arxiv.org/abs/2607.01506 (to appear in Proceedings of the ACM on Human-Computer Interaction, CSCW 2026)

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