Comment on FR Doc # 2026-12205

RE: Comment on General Services Acquisition Regulation; Acquisition of Information and Communication Technology — Large Language Model AI Safeguarding I am submitting this comment as a product designer and inclusive design researcher with experience in accessibility, inclusive design, and AI-mediated user experiences. I am completing a Master's Research Project at OCAD University's Inclusive Design Research Centre titled Who Gets to Use AI? Inclusive Prompting from the Edges, under the supervision of Dr. Jutta Treviranus, Director of the Inclusive Design Research Centre. My research examines how neurodivergent users interact with AI prompting systems, with attention to barriers affecting users with ADHD, autism, depression, and anxiety. The proposed GSAR clause addresses data safeguarding in the federal government's acquisition of large language model AI agents. I am writing to flag a related gap the current draft does not currently address: supporting the needs of users cognitive accessibility. Federal agencies are procuring and deploying LLM AI agents across contexts including education, employment, public welfare and benefits, health care, and government services. These systems are increasingly the interface through which the public, including people with disabilities accesses federal programs. AI agents are not cognitively neutral or unbiased. They assume users can formulate prompts in particular ways, interpret probabilistic outputs, detect errors, deconstruct complex tasks, and escalate when a system fails or outputs problematic information. For many neurodivergent users, those assumptions create barriers. They are not individual limitations, but design failures that procurement requirements can address before rollout to the public. Federal procurement is one of the most effective intervention points. If cognitive accessibility is not a condition of federal AI/LLM contracts, agencies will acquire and deploy AI systems that systematically exclude a significant portion of the public they are meant to serve. Retrofitting accessibility after deployment is significantly more costly and less effective than requiring it as a condition of acquisition. Recommended Considerations for the GSAR Clause: Recommended Considerations for the GSAR Clause: 1. Require cognitive accessibility standards as a condition of federal LLM AI procurement. Vendors supplying LLM AI systems to the federal government should demonstrate that their systems meet cognitive accessibility standards, including for users with ADHD, autism, depression, anxiety, and other cognitive disabilities. 2. Require vendor disclosure of known accessibility limitations. Vendors should disclose known limitations in their LLM systems' accessibility for users with cognitive disabilities, including limitations in prompt support, output clarity, error recovery, and human fallback pathways. 3. Require accessibility testing with disabled users before federal deployment. Agencies procuring LLM systems for public-facing or employee-facing use should conduct accessibility and usability testing with people with cognitive, psychiatric, learning, and communication-related disabilities (among the most common disabilities overall) before deployment. 4. Align with Section 508 and emerging Access Board AI guidance. The GSAR clause should explicitly align LLM and AI procurement requirements with Section 508 of the Rehabilitation Act and with emerging guidance from the U.S. Access Board on AI and disability. Fragmented requirements complicate compliance, but alignment ensures AI safeguarding and accessibility requirements reinforce each other. 5. Require accessible human fallback as a procurement condition. Any LLM/AI system procured for public-facing use cases should include an accessible pathway for users to reach humans for assistance or alternative communication when the AI fails them. This is especially critical for neurodivergent, elderly and low literacy users who may have greater difficulty detecting and recovering from AI failures. I would welcome the opportunity to contribute research findings to any follow-up process related to accessible federal AI procurement. Thank you, Tolu Adegbite Product Designer and Inclusive Design Researcher tolu.c.adegbite@gmail.com