AI policy in schools is being written by the wrong people. Building codes are never drafted without architects, yet we're writing the future of education without consulting the experts on accessibility. Too many school districts are developing AI policies without input from education professionals, handing the work to external consultants with no special education expertise. More than 30 states have rolled out K-12 AI guidance. Fewer than 70% of those documents even mention students with disabilities. The students who stand to gain the most from AI are invisible in the policies dictating its use. What follows draws from my LinkedIn series on AI in special education and from sessions at CASE 2025 — the legal foundations, the tools that already exist, and the choices districts are making right now. Most people in edtech don't know this: Congress already mandated that schools consider assistive technology for every student with an [[IEP::IEP (Individualized Education Program): the legally binding document under IDEA that specifies a student's disability-related educational goals, services, placement, and accommodations. Schools must develop and update it annually with input from a team that includes the family.]]. Not as an option — as a legal requirement under IDEA. And yet research consistently shows that AT consideration is either skipped or treated as a checkbox exercise. A building code that requires ramps, but nobody checks whether the ramps got built. AI changes this. Adaptive reading tools, automated writing feedback, real-time translation for families, content differentiated to any reading level in seconds. The gap is no longer capability. It's will. If your district is still debating whether to "allow" AI, the better question is: can you afford not to, when the law already expects you to deliver what these tools now make possible? The strongest argument for AI in special education isn't innovation. It's compliance. CASE 2025 foregrounded the hard questions. Does AI support or risk compromising a Free Appropriate Public Education? Can it adapt content, methodology, or delivery to meet a student's unique needs? How is it being used to collect and analyze data for progress monitoring? These aren't hypothetical. Project AI SCORE, developed at the University of Kansas, is a free system that instantly scores student writing across content, organization, and style on a six-point rubric, then returns personalized feedback. A teacher with 25 students might take a week to return writing feedback. For a middle schooler with a learning disability, that delay is the difference between connecting effort to outcome and losing the thread entirely. Think about getting your blood pressure checked once a year versus wearing a monitor that alerts you in real time. The information isn't just faster. It changes what you believe is possible. Frequent, immediate feedback lets students see their own growth while they're still in the act of growing. For a kid who's internalized a story about what they can't do, that evidence — right there, right then — rewrites the narrative. CASE 2025 raised a question most districts haven't answered: are staff trained to tell the difference between AI as an accommodation and AI as instruction? Can AI tools count as valid accommodations under a student's IEP or 504 Plan? The Oregon Department of Education's K-12 AI guidance offers a useful image: AI as an electric bike, not a robot vacuum. On the bike, the rider stays aware and in control while their effort is multiplied. The vacuum just does the work. AI can simplify complex text, personalize practice, and hand hours of administrative time back to teachers. It can also quietly privilege students with stronger devices, better home coaching, and more stable Wi-Fi. Every parent asks eventually: should I let my kid use AI? The sharper question is whether we're teaching them to think with it without letting it think for them. For neurodivergent students, this cuts both ways. These tools can personalize pacing, clarify complex text, and strip away counterproductive cognitive load. They can also erode productive struggle — the friction that builds mastery. A scaffold supports comprehension, organization, and expression. A crutch replaces synthesis, reasoning, and voice. The line between them is rarely technical. It's guided use, explicit norms, and shared reflection. When a district adopts an AI tool, someone should be asking: who owns the student data? Is the tool [[FERPA::FERPA (Family Educational Rights and Privacy Act): the federal law that governs access to and disclosure of student education records. It restricts how schools can share data with third parties, including edtech vendors.]]-compliant? Does it meet IDEA's confidentiality requirements? What stops it from encoding the biases already present in the training data? Most districts aren't asking. They're buying first. Project AI SCORE shows what responsible design looks like. Student writing never reaches researchers in identifiable form — it's deidentified before it leaves the district, and patterns are verified without anyone seeing the originals. The Student Data Privacy Consortium, with 28 state alliances backing its National Data Privacy Agreement, offers a governance model that other tools could follow. The NIST AI Risk Management Framework provides a broader evaluation lens. But having frameworks available isn't the same as using them. All of this is happening against a backdrop of declining student performance — NAEP 2024 showed eighth-grade science scores down across every percentile versus 2019, and PISA 2022 found nearly one in three 15-year-olds internationally below basic math proficiency. Special education teams are seeing more referrals where the issue isn't inability but interrupted opportunity. The urgency isn't theoretical. And the decisions that shape a student's trajectory don't happen at a desk. They happen in IEP meetings. These meetings have a problem that better classroom tools won't fix. Under IDEA, parents have the right to participate in the IEP process. But participation without comprehension is theater. For families navigating dense legal language, clinical terminology, or a second language, an IEP meeting can feel like being present but not included. Real-time translation of IEP documents. Plain-language explanations of procedural safeguards. AI-assisted preparation so families walk in knowing what decisions are on the table and what questions to ask. That's the gap between performative inclusion and the real thing. Most AI conversations in special education focus on classrooms. But the families with the least social capital benefit most from clarity and preparation — and they're the ones least likely to have it. Family-facing AI isn't novel. It's overdue. Of course, none of this works if the tools themselves aren't accessible. "Deliberate indifference" under Title II of the ADA means knowing about inaccessible digital platforms and failing to act. Every untagged PDF, every broken screen reader interaction, every platform that excludes a student with a disability leaves a trail. Organizations can't claim ignorance anymore. The legal landscape is messy. In Acheson Hotels v. Laufer (2023), the Supreme Court dismissed the case as moot without ruling on the merits. The national circuit split remains unresolved: the 1st, 4th, and 11th Circuits allow tester standing; the 2nd, 5th, and 10th reject it. Where you are determines what you face. For large public entities, the Title II ADA compliance deadline is April 24, 2026, though the DOJ submitted a revised interim final rule in February 2026 that may push this timeline. The underlying obligation doesn't change with the deadline. [[WCAG::WCAG (Web Content Accessibility Guidelines): the technical standard for digital accessibility, published by the W3C, that defines how web and software interfaces should be built to be usable by people with disabilities. Compliance with WCAG 2.1 Level AA is the benchmark referenced in most federal accessibility requirements.]] standards are one path to Title II compliance, but WCAG alone doesn't address accessibility for all students served under IDEA. Districts need to look past web standards to the full scope of digital tools used in instruction — the learning platforms, the assessment tools, the communication apps. Districts that treat AI as infrastructure — with clear norms around when it supports thinking and when it replaces it — will bend it toward inclusion. Districts that don't will watch it widen the gaps they meant to close. The tools already exist: the VITAL team at the University of Kansas has been curating free and low-cost AI tools for diverse learners, from Diffit for leveling content to MagicSchool for IEP supports to Goblin Tools for neurodivergent productivity. IDEA already requires districts to consider these kinds of supports. The question was never whether the tools would arrive. It's whether the adults in the room would be ready when they did. The rules are being written now. Without disability-informed voices in the room, protective frameworks turn into exclusionary walls — locking out the students who need these tools the most. Advocacy groups, special education researchers, legal specialists, and families need seats at the drafting table today. Not after the frameworks are finished. Now.
