Every Ohio public school district must adopt an artificial intelligence policy by July 1, 2026. Adopting the state’s model policy will make a district compliant. It will not, by itself, make the district safe. This analysis is written for the people who have to live with that gap: the administrators and boards who adopt the policy, the educators and special educators who apply it day to day, and the parents whose children are governed by it. The gap begins, and is widest, in special education.
Key Takeaways
- Ohio Revised Code 3301.24, enacted in the House Bill 96 budget, requires every Ohio school district, community (charter) school, and STEM school to adopt an AI policy by July 1, 2026. The statute mandates that a policy exist. It does not dictate what the policy says.
- A district can adopt the Department of Education and Workforce (ODEW) model verbatim and be compliant. The model is adequate and a sensible starting point. It is also silent on many of the questions most likely to generate liability, functioning more as a baseline checklist than a complete governance framework.
- The sharpest gap is special education. A binding decision of the federal appeals court covering Ohio holds that AI tools which do a student’s work for them can deny that student a free appropriate public education. The model never draws the line that case turns on: the difference between AI that remediates a disability and AI that merely masks it. A student can be passing every class while the district is quietly accumulating liability.
- The other exposures the model leaves to the district: student data contracts, AI detection software used for discipline, and student safety, an area where even seasoned commentators misstate the governing law.
The Deadline Is the Easy Part
House Bill 96, Ohio’s 2025 biennial budget, added Section 3301.24 to the Revised Code and set two dates. ODEW had to publish a model AI policy by December 31, 2025, which it released in early January 2026. Every public school district, community (charter) school, and STEM school must adopt a policy on the use of artificial intelligence by July 1, 2026. Ohio is the first state to convert AI guidance from a suggestion into a statutory duty.[1]
A careful read of the statute yields a few important points. The statute mandates that each district adopt a policy. The model policy is merely that: an adequate baseline. The statute does not say what the policy must contain, does not require teaching AI, and does not require using AI. A district may adopt ODEW’s model verbatim or write its own.[2] If a board wants to be compliant by July 1 and nothing more, the route is short: adopt the model as written. It is sufficient on its face.[3]
Compliance, in other words, is the floor. Everything worth reading past this point is the part the statute and the model leave to the district.
The Blind Spot: Special Education and the William A. Trap
The first place a district policy will be tested is special education, and there is little guidance on the point. One case provides real clarity, and it is binding here. In William A. v. Clarksville-Montgomery County School System, the U.S. Court of Appeals for the Sixth Circuit, whose decisions bind federal courts in Ohio, Kentucky, Michigan, and Tennessee, affirmed that a dyslexic student who graduated with a 3.4 grade point average had been denied a free appropriate public education. He had passed by routing every assignment through a chain of tools: he dictated a topic using speech-to-text software, pasted it into an AI program like ChatGPT to generate the paper, then ran the result through Grammarly. The accommodations, the court found, simply did the work for him. They masked his inability to read rather than teaching him to read.[4]
The distinction the court drew is the one every Ohio district should build into its policy: the difference between an accommodation that remediates a disability and one that merely masks it. Specially Designed Instruction under the Individuals with Disabilities Education Act has to do the first. AI makes the second effortless, and therefore easy to mistake for progress. Transcription and dictation tools have always done some of this masking, just imperfectly enough that no one confused the output with mastery. Generative AI does it well enough to fool a transcript. A student who produces fluent essays with these tools can look like he is thriving while learning nothing, and a strong GPA is no defense if the record shows the technology, not the student, was doing the work.
To be clear, the holding does not ban assistive AI. Used deliberately, this technology can be genuinely liberating for students with disabilities. The holding is an argument for intent and documentation. For any student on an IEP or 504 plan, the IEP team should be able to show that AI supports skill-building rather than replacing it, that the team chose the tool deliberately, and that someone is measuring whether the underlying skills are improving. Federal regulators identified the same ground: in November 2024, the U.S. Department of Education’s Office for Civil Rights issued guidance cautioning that AI used in services such as developing IEP or Section 504 documents can raise disability discrimination concerns. The Department has since rescinded that guidance, but the statutes it interpreted remain fully in force, and William A. remains binding law.[5] The ODEW model, by contrast, addresses special education only by seating a special education representative on its AI workgroup and suggesting that districts review related policies. That is the kind of gap that surfaces in a due process complaint rather than a compliance audit.
The debate is not new, only escalating. Ohio’s own cell phone law tracks the same shape. Revised Code 3313.753, as amended by House Bill 96, required every district to prohibit student cell phone use during the instructional day by January 1, 2026, with carve-outs for students whose IEPs or 504 plans require a device or who need one to monitor a health condition. Every district enforcing that ban is already making the remediate-versus-mask call for phones: when is a device an accommodation that builds independence, and when is it a workaround that substitutes for the underlying skill? Generative AI is the next iteration of the same question, with higher stakes. A phone can transcribe a lecture; ChatGPT can write the paper. A district policy that handles phones thoughtfully and AI mechanically will fail in the same way, for the same reason.
What Else the Model Leaves to the District
Student Data: The Contract Is the Control
Any tool that handles student information in an educational setting is already governed by the Family Educational Rights and Privacy Act (FERPA). FERPA lets a school share student records with a vendor only when the vendor performs an institutional service under the school’s direct control and uses the data only as directed. The standard terms of service for consumer chatbots are weak on confidentiality and do not meet that standard. Ohio sharpened the point in Senate Bill 29, which provides that education records handled by a technology provider under contract with a district are solely the property of the district and bars the provider from selling, sharing, or using those records for commercial purposes, subject to narrow statutory exceptions.[6]
In practice: a teacher who pastes a student’s IEP or a behavioral note into a free public AI tool may be handing protected data to a vendor with no contract and no control. A workable policy names approved tools, forbids the rest for anything involving student data, and requires vendor terms that bar training on district data, return or delete records at termination, and provide breach notice. Some districts now steer staff to AI services covered by a K-12 privacy and data processing agreement, with Microsoft 365 configured for an education tenant a common example, and decline to authorize consumer chatbots like ChatGPT for anything involving student information. Columbus City Schools, which adopted its AI policy in March 2026, took exactly this approach.[7]
AI Detectors – Are They Evidence of Anything?
The reflex to buy AI detection software and treat a high score as proof of cheating is common, and it is an unreliable basis for discipline. Detectors assign probabilities, not findings. They flag writing that is statistically predictable, which means they disproportionately flag non-native English speakers, neurodivergent students, and anyone with a clean, formulaic style. A Stanford-led study published in the journal Patterns found that detectors misclassified more than half of TOEFL essays by non-native writers as AI-generated, an average false positive rate of 61.3 percent. The University of Pittsburgh disabled Turnitin’s AI detector rather than carry that risk.[8]
A policy that rests discipline on a detector score invites both wrongful accusation disputes and civil rights exposure, and the model says nothing about it. Detection tools carry independent risks, including the structural awkwardness of using AI to flag AI in classrooms where AI-assisted grading is already in use. Process-based assessment, meaning drafts, revision history, and in-class writing samples, produces a defensible record where a detector score does not.[9]
Student Safety: Know Which Law Actually Applies
Generative tools have made it trivial to fabricate explicit images of a classmate, so districts are right to address synthetic intimate imagery in policy. The model cross-references Braden’s Law, and commentators routinely miscast that statute as an AI or deepfake law. It is not. Braden’s Law, House Bill 531 of the 135th General Assembly, signed January 8, 2025, is a criminal sextortion statute. It makes sexual extortion a felony and contains no AI or deepfake provisions.[10] The laws that actually reach AI-generated intimate images are the federal Take It Down Act, which creates criminal liability for publishing nonconsensual intimate imagery, including AI-generated “digital forgeries,” and requires covered platforms to remove such content under a notice-and-takedown process that took effect in May 2026, and Ohio Senate Bill 163, which targets AI-generated child sexual abuse material and deepfakes and, as of this writing, has passed the Ohio Senate and is pending in the House.[11] The practical move for a district is to classify creating or sharing synthetic intimate imagery as a serious disciplinary offense and route incidents to law enforcement and victim support rather than the help desk.
What the Model Does Well, and What It Leaves Undone
To be clear, the ODEW model is a solid document, best understood as a starting point. It rejects the idea that AI replaces teachers, requires districts to vet vendors and protect personally identifiable information, sets academic integrity expectations, and stands up a standing AI workgroup. Adopting it is a defensible way to meet the deadline. The point is narrower and more useful: the model is largely silent on the questions that produce liability. It does not draw the remediate-versus-mask line for special education, says nothing about AI detection software, does not address teacher labor or employment, treats equity and algorithmic bias only obliquely through procurement, and carries no enforcement mechanism. Verbatim adoption is a reasonable start and an incomplete finish.
How a District Adopts, and Later Amends, the Policy
A board’s authority to adopt rules for the governance of its schools comes from Revised Code 3313.20, and any board action, including policy adoption or amendment, must occur in open session under Ohio’s Open Meetings Act.[12] The two-reading process most Ohio boards use, a first reading to introduce a policy, a second reading after a window for review and public comment, and a vote, is a matter of board bylaws and common practice rather than statutory mandate. Many districts run this through a policy service such as NEOLA.
The consequence worth planning around is that amending a policy is the same board action as adopting one, not an administrative tweak. A district that adopts the model verbatim by July 1 can revise it at the next reading cycle, and given how fast the tools and the law are moving, it should. A sensible sequence: adopt or customize by July 1; convene the workgroup the model contemplates; and calendar the first amendment cycle for the 2026-27 school year.
A Note for Parents
For parents, the policy your district adopts this summer is worth reading for three reasons. It tells you which AI tools your child may use and what happens to their data. It tells you how your child’s teacher will treat AI in graded work, which matters if your child is ever accused of cheating on the strength of a detector score. And if your child has an IEP or 504 plan, William A. is your leverage: your child is entitled to instruction that builds skills, not tools that hide their absence. Ask the district how it tells the two apart.
Frequently Asked Questions
Does every Ohio school really have to adopt an AI policy by July 1, 2026?
Yes. Revised Code 3301.24 requires every school district, community school, and STEM school in Ohio to adopt a policy on the use of artificial intelligence by July 1, 2026.
Does the law tell districts what to put in the policy?
No. It requires that a district adopt a policy. The content is left to local control, which is why the state model is a starting point rather than a safe harbor against everything that can go wrong.
Can a district just adopt the state model policy?
Yes, and doing so makes the district compliant. But the model is largely silent on special education, AI detectors, vendor contracts, and labor, so verbatim adoption leaves real exposure in place.
What does William A. mean for special education students?
In the Sixth Circuit, which includes Ohio, AI accommodations cannot substitute for Specially Designed Instruction. If a student is passing only because AI is doing the work, the district may be denying a free appropriate public education, regardless of the student’s grades.
Is AI detection software a safe basis for student discipline?
No. Detectors produce probabilities, not proof, and carry documented bias against non-native and neurodivergent writers. Process-based assessment and human judgment produce a defensible record; a detector score alone does not.
How does a district change the policy later?
The same way it adopts one: a board action through the reading process in open session. Plan on an annual review, because both the tools and the law are moving quickly.
How KJK Can Help
KJK works with organizations of all types to develop, implement, and operationalize AI policies designed to limit liability, comply with the law, and be clearly understood by the people they govern. If your board has adopted the ODEW model, or has a policy of its own, we will review it against the exposures above and current Sixth Circuit law and tell you where it leaves the district open. If your board needs a policy built for its actual tools, contracts, and students, we will develop one with you and help walk it through the reading process. The goal either way is a policy that does more than clear a deadline.
For more information or an initial consultation, contact Ted Theofrastous (TCT@kjk.com), Chair of KJK’s AI Practice.
This article is for general information and is not legal advice. For guidance on a specific district’s policy or situation, consult counsel.
—-
Footnotes:
[1]Ohio Rev. Code § 3301.24, enacted by Am. Sub. H.B. 96, 136th Gen. Assemb. (Ohio 2025), eff. Sept. 30, 2025; Ohio Dep’t of Educ. & Workforce, AI Model Policy for Ohio Districts and Schools (released Jan. 2026).
[2]Ohio Rev. Code § 3301.24(B) (a district or school “may adopt the department’s model policy”). The statute requires adoption of a policy; it does not prescribe content, mandate AI instruction, or require AI use.
[3]Ohio Dep’t of Educ. & Workforce, AI Model Policy for Ohio Districts and Schools (2026) (acceptable-use, privacy/FERPA, vendor-evaluation, academic-integrity, AI-workgroup, and periodic-review provisions).
[4]William A. v. Clarksville-Montgomery Cnty. Sch. Sys., 127 F.4th 656 (6th Cir. 2025) (Kethledge, J.), applying Endrew F. v. Douglas Cnty. Sch. Dist. RE-1, 580 U.S. 386 (2017). The decision binds districts in Ohio, Kentucky, Michigan, and Tennessee. The AI workflow was one of several accommodations the court addressed.
[5]U.S. Dep’t of Educ., Office for Civil Rights, Avoiding the Discriminatory Use of Artificial Intelligence (Nov. 2024) (identifying disability-discrimination risks, including AI used to develop IEP or Section 504 documents). The Department has since formally rescinded the guidance; the statutes it interpreted remain in force.
[6]20 U.S.C. § 1232g; 34 C.F.R. pt. 99 (FERPA “school official” exception and direct-control requirement); Ohio Rev. Code § 3319.326 (enacted by Sub. S.B. 29, 135th Gen. Assemb., eff. Oct. 24, 2024, as amended).
[7]Columbus City Schools, Board Policy 7540.09 (adopted Mar. 31, 2026); Columbus City Schools, Learning Technology (Microsoft Copilot designated the district’s primary AI service; no agreement permitting consumer ChatGPT for student data).
[8]Weixin Liang et al., GPT Detectors Are Biased Against Non-Native English Writers, 4 Patterns 100779 (2023) (61.3% average false-positive rate on TOEFL essays by non-native writers); Univ. of Pittsburgh, Univ. Ctr. for Teaching & Learning (disabling Turnitin’s AI-detection feature).
[9]Ohio Dep’t of Educ. & Workforce, Integrating Artificial Intelligence in Content-Area Instruction (the tiered “Student AI Usage” continuum appears in this instructional resource, not in the adoptable model policy).
[10]Am. Sub. H.B. 531, 135th Gen. Assemb. (Ohio 2024) (“Braden’s Law”), signed Jan. 8, 2025, eff. Apr. 9, 2025; codified in relevant part at Ohio Rev. Code § 2905.111. The act criminalizes sexual extortion and contains no AI or deepfake provisions.
[11]TAKE IT DOWN Act, Pub. L. No. 119-12 (signed May 19, 2025) (criminal provisions effective on enactment; covered platforms’ notice-and-removal obligations effective May 19, 2026). Ohio’s S.B. 163, 136th Gen. Assemb., addressing AI-generated child sexual abuse material and deepfakes, passed the Senate unanimously on May 20, 2026, and is pending in the House as of this writing.
[12]Ohio Rev. Code § 3313.20 (board authority to adopt rules); Ohio Rev. Code § 121.22 (Open Meetings Act).