AI systems increasingly shape how people learn, work, travel, and access healthcare and pub-lic services. Yet, people who could benefit most often face barriers to using them. For people with disabilities, older adults, and people with limited technical literacy, AI raises challenges that existing accessibility guidelines do not fully address. AI systems can change their behav-ior, produce variable outputs, and make decisions that are difficult to understand or challeng-ing. Since June 2025, the European Accessibility Act has required covered products and ser-vices to meet accessibility requirements. Germany’s BFSG implements the Act, while the EU AI Act adds transparency obligations for certain AI systems. A central challenge remains: how can we design AI systems that are both accessible and explainable in everyday use?

This Dagstuhl Seminar brings together researchers from human-computer interaction, acces-sibility studies, AI and explainable AI, and cognitive science with practitioners from industry. We look at accessible human-AI interaction from two sides. The first is access: making AI-powered systems usable for people with diverse sensory, motor, and cognitive abilities. The second is comprehension: moving explainable AI beyond explanations designed for data sci-entists, toward explanations that everyday users can understand, question, and act on. Much current work uses AI to build assistive tools; we focus on the complementary problem of AI systems that create new barriers of their own. How can a passenger with a cognitive disability navigate AI-powered transport routing? How can older adults meaningfully take part in de-signing the healthcare AI that affects their treatment? How do we communicate model uncer-tainty to someone who cannot see a confidence chart?

The seminar is organized around six topics:

• foundations and frameworks: extending guidelines such as WCAG to AI;
• interaction beyond the screen: voice, haptic, sonified, tangible, and mixed reality;
• participatory, inclusive AI development ("Nothing About Us Without Us");
• applications in public services, healthcare, education, and assistive technology;
• infrastructure: adaptive explanation algorithms, user modelling, datasets, benchmarks;
• social, ethical, and policy questions of bias, paternalism, and global inequality.

Instead of conventional presentations, participants introduce their work in short ignite talks, and most of the week goes to hands-on exploration and collaborative working groups. Our aim is to establish accessible human-AI interaction as a research area in its own right. We plan to produce a shared research agenda, develop evaluation frameworks and design guidance for AI systems, translate regulatory requirements into practical recommendations, and start joint publications and follow-up activities. We invite you to help shape AI that is designed for the full spectrum of human diversity.

Creative Commons BY 4.0

Giulia Barbareschi, Andreas Holzinger, Kai Kunze, Yvonne Rogers, and Anusha Withana