Addressing the Void of AI Policies in Education for Students With Specific Learning Disabilities

Synthesis: Shin et al. (2026) identify the current state of AI policies in U.S. education for students with specific learning disabilities (SLD) and propose actionable policy recommendations. Combining Large Language Models (LLMs)-based topic modeling (structural topic modeling + Sentence-BERT embeddings) with two rounds of Delphi surveys of 17 experts, they analyze 12 U.S. AI-in-education policy documents (2015–2025) — finding that only 2 of 12 specifically address learning disabilities and that 18 policy topics present in other-disability or general AI policy (risk assessment, data protection, legal risk management, ethical guidelines, higher-education AI) are missing from SLD policy. The study produces 36 validated policy items across five themes — inclusive and Personalized Learning, ethics/equity/inclusion, student empowerment and AI literacy, assessment and research, and educator preparation — with student empowerment and AI literacy ranked the most essential by experts. Grounded in the U.S. Assistive Technology Act (2004) and IDEA (2004), the paper frames AI as both assistive technology and a source of FAPE-related, privacy, equity, and accessibility risks, calling for AI policies that are evidence-based, accessible, and inclusive of neurodiverse and multilingual learners.

Core Finding

There is a void of AI-specific policy for students with specific learning disabilities (SLD) in U.S. education: of 12 analyzed AI-in-education policy documents, only 2 specifically address learning disabilities, and key policy topics (AI risk assessment, data protection, legal risk management, ethical guidelines) covered for other disability groups and general AI education are missing from SLD policy. Experts converge on student empowerment and AI literacy as the most essential policy priority, followed by inclusive and personalized learning and educator preparation — with ethics, equity, inclusion, and evidence-based assessment as supporting pillars.

The policy void and its legal context

Federal law — the U.S. Assistive Technology Act (2004) and the Individuals with Disabilities Education Improvement Act (IDEA, 2004) — mandates evidence-based practices and assistive-technology evaluation within each student's Individualized Education Program (IEP) and requires a free, appropriate public education (FAPE). Yet despite rapid AI adoption, no formal, evidence-based guidelines exist to help educators, practitioners, students, or families implement AI for learners with SLD. The study's document-level similarity analysis found SLD policies closer to other-disability policies (M = 0.50) than to general AI policies (M = 0.42), and identified 18 topics present elsewhere but absent from SLD policy. The case W.A. v. Clarksville/Montgomery County School System (2024) — in which a student was given AI as an accommodation yet graduated unable to read, with the court finding FAPE was not provided — illustrates the legal and ethical stakes of deploying AI as an accommodation without adequate instruction.

Five themes of validated AI policy recommendations

The 36 final policy items cluster into five themes:

1. Inclusive and personalized learning (11 items, 30.56%) — AI for multiple ways of knowing, personalized/differentiated resources, neurodiverse- and ELL-specific tools, AI-driven literacy tools, OCR and read-aloud accessibility, and instructional models centered on individual SLD needs.
2. Ethics, equity, and inclusion (9 items, 25.00%) — inclusive data representing SLD, co-design with SLD students, equity impact evaluation, data privacy protection, government risk-management guidelines, and accessibility prioritized in AI development principles.
3. Student empowerment and AI literacy (6 items, 16.67%) — Teaching students with SLD to use AI responsibly, ethically, and independently; avoiding overreliance and plagiarism; building Self-Efficacy and Prompt Engineering skills.
4. Assessment and research (6 items, 16.67%) — evidence-based, accessible, and adaptable AI Assessment tools; digital accessibility standards; funding for SLD AI research; mandates grounded in rigorous research.
5. Educator preparation (4 items, 11.11%) — AI literacy in teacher-preparation programs, accessible school–parent communication, leadership training, and an ethical statement as a first step for both general and special education.

Accessibility vs. accessible learning

This paper usefully illustrates the distinction the knowledge base draws between narrow Accessibility (captions, alt text, Assistive Technology, accommodations, OCR, read-aloud, customizable settings — the concrete "access features" called for across multiple items) and the broader Inclusive Learning (multiple ways of knowing, personalized and universal design for learning, differentiated instruction, neurodiverse and multilingual inclusion). The policy recommendations span both: specific accessibility accommodations for tools and assessments sit alongside wider demands for inclusive, personalized, and universally designed learning experiences.

AI literacy, equity, and the digital divide

Experts prioritized teaching AI literacy to students with SLD as an empowerment and Equity measure, while flagging risks of overreliance, plagiarism, and loss of Critical Thinking skills. The paper also highlights the Digital Divide (cost of advanced AI tools exacerbating disparities between wealthy and impoverished schools), Ethics and bias in algorithms, and teacher-training gaps — underscoring that equitable AI benefit for SLD students requires policies addressing bias, privacy, and professional development.

Relevance to the knowledge base

This paper is a cornerstone reference for the knowledge base's accessibility/disability topic, directly contributing to Educational AI Policy (a concrete, validated set of AI policy recommendations), Special Education (SLD-specific needs and IEP/FAPE legal grounding), Inclusive Learning (the broad inclusion agenda), and the emerging Accessibility concept (specific assistive/accommodation features). It provides an evidence base for how AI policies must be written to serve students with disabilities, offers a Human-in-the-Loop methodology exemplar (LLM topic modeling + expert Delphi validation), and connects accessibility to ethics, equity, teacher preparation, assessment, and AI Literacy.

What this means for practice

• Policymakers. Close the SLD gap explicitly. Only 2 of the 12 U.S. AI-in-education policy documents analyzed address learning disabilities, so write SLD-specific provisions rather than assuming general AI policy covers these students.
• Policymakers. Import the topics other disability and general AI policies already cover — AI risk assessment, data protection, legal risk management, and ethical guidelines — into SLD policy, where the study found them consistently absent.
• Administrators. Start implementation with the theme experts ranked most essential: student empowerment and AI literacy, taught through explicit instruction, guided practice that builds Self-Efficacy, and clear rules on responsible use and plagiarism avoidance.
• Administrators. Require that any AI written into an IEP or issued as an accommodation carries documented instruction and monitoring, since the paper ties unaccompanied deployment of AI accommodations directly to FAPE-related legal and ethical risk.
• Teacher educators. Add AI literacy to both general and special education preparation programs, and make an institutional ethical statement the first, lowest-cost step toward policy compliance.

Limitations

• The Delphi panel was small and shrank across rounds: 23 experts were invited, 17 completed Round 1, and only 14 completed Round 2, so the 36 ranked policy items rest on those who stayed in the process.
• Panelists were recruited purposively through academic networks, professional associations, and organizations against five qualification types, so priorities reflect that selected group's expert judgment rather than measured outcomes for students with SLD.
• The document analysis covered 12 publicly available U.S. policy documents from a Westlaw search spanning January 2015 to March 2025; non-U.S. and non-public documents were out of scope, and only 2 of the 12 documents were SLD-specific.
• Item generation was machine-assisted — structural topic modeling plus Sentence-BERT embeddings produced candidate items that researchers then manually reviewed and revised — and the SLD-specific policy corpus was thin, which constrains how much topic evidence those items can rest on.

Citation

Shin, M., Deniz, F., Watson, L., Dieterich, C., Ewoldt, K. B., Johnson, F., Kong, J. E., Lee, S. H., & Whitehurst, A. (2026). Addressing the void of AI policies in education for students with specific learning disabilities. Learning Disability Quarterly, 49(3), 134–146.