Who Sets the Rules for Classroom AI

The policy chain starts above the district, but it does not end there. The federal executive order of April 23, 2025 promotes AI literacy and federal education initiatives. It does not approve an individual classroom chatbot. Federal protections for student records still apply. State lawmakers and agencies can then set duties for policy, notice, training or oversight. Districts make product and account decisions within those protections, while teachers set the terms of an assignment. ^{1 11 4 8 34}

Illinois shows why the kind of state document matters. Its law required the state board to issue guidance by July 1, 2026, and guidance appeared in June. That provision did not directly require each district to adopt an AI policy. An illustrative resource marks consumer ChatGPT accounts “Not Approved” in a student-data scenario. The resource also disclaims statewide endorsement or requirements for the listed tools. Turning that example into a statewide product ban would change its meaning. ^{2 3}

Ohio requires a different action. Traditional districts, community schools and STEM schools had to adopt a formal AI policy by July 1, 2026. The state model offers sample language about privacy and student work. Those suggested terms do not prove that every local board adopted the same policy. For a student, the consequential document is the version adopted by the school. For a teacher, that document connects the state requirement to the actual assignment and tool rules. ¹⁶

Idaho also requires local action, with districts and public charter schools directed to align their generative-AI policies with a statewide framework. Its law took effect July 1, 2026. It directs the state education department to prepare guidance for parents and guardians. Receiving that guidance and consenting to a child’s use are different events. A family still needs the local policy to understand approved tools and classroom conditions. The framework supplies a starting point for that inquiry. ¹⁷

California and Massachusetts illustrate the limits of treating guidance as a uniform requirement. California’s June 2026 model expressly says compliance is not mandatory. Massachusetts says its guidance neither recommends nor requires schools to use AI. Both can inform local choices about privacy, oversight and academic integrity. Their existence does not show which district adopted which safeguards. A reader comparing jurisdictions needs to know whether a document requires an action, recommends one or records a local choice. ^{5 6 3545}

Family choice varies even among documents that address it directly. Houston requires advance written consent for qualifying access by students in eighth grade and above. Oklahoma’s statute permits parents or guardians to decline student-facing AI in writing without an academic penalty. It also calls for educator-directed use and annual written disclosure to parents. A parent seeking a choice faces different procedures in these two settings. One gives advance permission. The other uses a written opt-out. ^{34 38}

Oklahoma’s local-policy requirement applies before the 2027–28 school year. Families should be able to identify the disclosure and the written opt-out channel in the applicable policy. ³⁸⁴⁶

Other consent provisions cover narrower circumstances. Utah’s statute requires written parental consent for specified AI sandbox courses. That does not establish the same requirement for every ordinary classroom use. California’s optional model recommends annual consent for specified AI-driven data collection or autonomous interactions. The scope of the activity and the authority of the document are part of the family choice itself. ^{5 6 2047}

Seattle offers a different mechanism through its network rules. A family can request a restricted network group, but that is a general network choice rather than an AI-specific opt-out. The district’s 2024–25 AI pilot involved five high schools and two middle schools. That limited, dated pilot does not describe current access for every Seattle student. A family looking for today’s conditions needs the current district procedure and the scope of the requested network restriction. ^{10 12}

Device and account rules add further distinctions. Fairfax’s guide says the district does not provide or approve student general-purpose generative AI on district-issued devices. For personally owned devices, students should first check the teacher’s assignment rules and ask a caregiver before creating an account. DCPS includes approved student-learning tools in its acceptable-use policy. Its annual enrollment acknowledgement records receipt of the policy. That acknowledgement alone cannot be reported as a separate AI consent form. ^{14 26}

Boston’s cited policies call for privacy impact assessment, approved tools and family information. Those steps can help a family understand the district’s choices, but information and permission serve different purposes. A useful explanation names the tool, the eligible student and the applicable choice procedure. It also identifies who can answer a question about data or account access. Families should be able to locate those details without assuming that a general technology notice resolves every AI decision. ^{13 23 34 38}

A second set of limits concerns the decisions schools make about students. New York City’s cited guidance reserves grading, placement and discipline for qualified humans. That boundary reaches beyond whether a student can use a chatbot to help with work. Product approval also carries a student-data safeguard: vendors cannot use that data to train models. A family evaluating a tool needs both pieces. Permission to use a service and authority to make a consequential decision are separate questions. ⁸

Utah’s law takes a related but distinct route through a required state model. That model must prohibit students from using AI on academic assignments unless an educator authorizes the specific assignment. It must also bar educators from independent AI grading or high-stakes determinations. The state model is due before December 1, 2026, and local education agencies must base policies on it. These prescribed terms do not establish that every agency has already adopted its completed policy. ²⁰

Virginia requires school boards to establish, implement and enforce policies consistent with state guidance. Its framework addresses vendor use of student information for model training, human judgment in consequential decisions and teacher access to approved instructional AI transcripts. The transcript provision makes oversight more specific than a general promise that adults remain involved. It identifies information a teacher can review when AI supports instruction. The cited framework still needs to be read alongside the board’s applicable local policy. ¹⁹

Academic misconduct is another decision where the distinction matters. Chicago’s student rules call for investigation of suspected AI misuse. Suspicion alone does not become an automatic conclusion in that process. California’s optional model says AI-detector output should not be the sole basis for punishment or a grade penalty. These provisions address how a school reaches a judgment. A list of approved tools, by itself, cannot explain the procedure a student faces when a school questions work. ^{4 6}

Product approval is also more specific than a general statement that schools allow AI. Chicago directs students to its vetted EdTech Catalog and teacher permission. Philadelphia names Google Gemini and Adobe Express with Firefly as approved tools. Its guidance treats unapproved tools used for district work as an acceptable-use violation. At Philadelphia’s Masterman school, the cited handbook adds a continuum of teacher discretion. That school-level detail cannot be applied to every Philadelphia classroom. ^{4 41 42}

The next decisions will arrive on different schedules. North Carolina’s enacted framework calls for a state model by December 31, 2026, followed by distribution by January 15, 2027. Public school units must adopt policies by June 30, 2027. Specified teacher professional development follows by June 30, 2028. These are four stages with different actors and purposes. The model deadline does not establish completed local policies or completed teacher training. ^{39 4348}

Maryland uses a relative deadline. Its law took effect June 1, 2026 and requires local school systems to issue a policy within 120 days after the state releases specified guidance. The law also requires an AI coordinator and addresses AI literacy standards by June 1, 2027. A local calendar needs the guidance’s actual release date as well as the statute. ³³⁴⁹

Utah’s dates also need to remain separate. Its state AI model is due before December 1, 2026. A distinct balanced-technology policy requirement has a July 2027 deadline. Those obligations address related school technology decisions, but the later date does not replace the earlier AI-model duty. Families asking what applies to an assignment need the relevant AI policy. School leaders planning adoption need to identify which policy each date governs before using it in a calendar. ²⁰

Tennessee combines an existing local-policy duty with later teacher training. Public Chapter 550 required district boards and charter governing bodies to implement AI-use policies by the 2024–25 school year. Those policies cover students, teachers and staff. Public Chapter 1056 sets a separate training deadline for teachers of grades six through 12. It is August 1, 2028, or within two years of licensure if that is later. The later training date does not delay the earlier policy requirement. ^{25 27}

Florida shows how quickly a dated category can become incomplete. The August tracker placed the state in its no-statewide-rule category. On September 16, 2026, the state education department announced the board’s adoption of AI rules and described parent notice and opt-in. That announcement records a change after the snapshot. ^{7 18 29 3650}

A proposed Florida amendment includes a July 1, 2027 deadline for local internet-safety policies, parent choice and transcript-retention language. The proposed date cannot yet be treated as an operative deadline from these sources. Families need the district’s current practice, while leaders need the filed final rule before treating those proposed terms as binding. ^{9 18 29 3651}