In April, the Educating All Learners Alliance and New America released the second version of their policy paper, Prioritizing Students with Disabilities in AI Policy. It was unveiled at the ASU+GSV Summit on April 14 and funded by the EDSAFE AI Alliance. Its argument is that students with disabilities should be considered at the start of AI policy design, not added later.
About 7.5 million students with disabilities attend U.S. schools. The paper reports that 57 percent of licensed special education teachers use AI to help develop IEPs or 504 plans, the documents that set out a student’s services and accommodations. It doesn’t cite a source for that figure.
The paper lists the risks it sees. One is bias: it cites Penn State research in which sentiment tools rated disability terms like “blind” and “deaf” as negative, and a 2025 review that rated none of 11 AI interventions for learning disabilities as low risk for bias. Others are the privacy of disability data, free tools that train on student data, interfaces that the students themselves can’t use, surveillance, and automated recommendations that crowd out individual decisions.
Much of the paper reads as a procurement checklist.
What it asks of school districts
- Require WCAG 2.1 AA conformance and an Accessibility Conformance Report from vendors.
- Require model cards that disclose a tool’s limitations and potential biases.
- Write contracts that cover data ownership, limits on use, and deadlines for deleting data. Block vendors from training on student data.
- Configure privacy settings by hand rather than accepting defaults.
- Include special education directors, advocates, and families in decisions to adopt a tool.
- Keep a public list of approved tools, and review them quarterly against set criteria for dropping one.
- Budget for audits and evaluation after the purchase, not just for the purchase.
What it asks of states
The paper wants states to issue guidance on how IDEA, Section 504, and the ADA apply to AI, and to provide model contract terms for disability data. It also asks states to require that a person, with parent input, make the final decision on placement, services, accommodations, and discipline.
These are recommendations from advocacy and policy groups, not legal requirements. One piece is already law for public schools: the ADA Title II rule that sets WCAG 2.1 AA as the standard for state and local government web content and apps. An AI tool a district puts in front of students is part of that.
A tool the students can’t use fails before anyone gets around to measuring its bias.
Sources