"Accessibility people, you go work on that thing of yours over there": Addressing Disability Inclusion in AI Product Organizations

Disability considerations are often siloed within AI organizations: The paper found that AI practitioners experience "friction when triaging problems at the intersection of responsible AI and accessibility practices," suggesting that disability inclusion is frequently treated as a separate concern rather than integrated into core AI development processes. This fragmentation can lead to disabled users being overlooked in product decisions.

Existing guidelines create conflicting requirements: Practitioners identified "contradictions between accessibility and responsible AI guidelines," meaning that following best practices in one area may conflict with best practices in another. This leaves teams uncertain about how to prioritize when these standards clash, potentially disadvantaging disabled users.

Critical data gaps about disabled users exist: The study found "gaps in data about users with disabilities," indicating that AI teams lack the information needed to understand how their systems affect people with disabilities. Without this data, responsible AI practices cannot adequately account for disability-related harms.

Informal networks, not formal structures, drive disability inclusion efforts: The researchers found that practitioners "gathered support for addressing the needs of disabled stakeholders by leveraging informal volunteer and community groups within their company." This suggests that disability inclusion depends on individual initiative rather than institutionalized processes, making it fragile and inconsistent.

Systemic changes are needed beyond individual effort: The paper calls for "new resources and process changes," indicating that relying on well-intentioned practitioners alone is insufficient—organizations need formal structures, guidelines, and resources dedicated to ensuring AI systems work for people with disabilities.