Rhetoric vs Responsibility: How Tech Companies Shape AI for Accessibility
Aparajita S Marathe, Quan Zhou, Achi Mishra, Anne Marie Piper · 2026 · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26) · doi:10.1145/3772318.3791854
Summary
Marathe, Zhou, Mishra, and Piper conduct a critical discourse analysis (CDA) of 126 public-facing blog posts and news articles published between 2016 and 2025 by 11 leading U.S.-based AI companies — Adobe, Amazon, Apple, Eleven Labs, Google, Meta, Microsoft, Nvidia, OpenAI, Salesforce, and Waymo — to examine how the tech industry frames its role in 'AI for accessibility.' The authors situate their work at the intersection of HCI, Organizational Studies, and Critical Disability Studies, drawing on scholarship from Shew on technoableism, Bennett and Keyes on disability justice, and Jackson and colleagues on the 'disability dongle' to interrogate corporate narratives that most accessibility research has treated uncritically. Using feminist disability studies and Fairclough's CDA as methodological anchors, the four authors collected articles via each company's native search, filtered out announcements and non-accessibility content, and performed iterative thematic coding. Rather than cataloguing AI products, the paper asks how corporate writing constructs participation, agency, and legitimacy. The analysis produces four overarching themes: the modes of disabled participation that companies render visible, the agency bestowed upon AI itself, the ways companies reinforce their role in shaping AI futures, and the discursive strategies used to legitimize AI-for-accessibility investment. The paper closes by proposing concrete moves toward more democratic AI futures — redistributing agency, questioning dominant values such as independence and productivity, and replacing symbolic involvement with collective accountability mechanisms like ProcureAccess and Disability:IN.
Key findings
The authors document a recurring pattern in which disabled people appear in corporate discourse as beneficiaries, sources of inspiration, product testers, co-creators, or data generators — but almost never as decision-makers with authority over product direction. Even when firms emphasise 'co-design' or open-source releases, participation is gated by corporate permission, and quotes such as Google's Project Euphonia recruitment (5,000 more speech samples needed) reveal an extractive relationship where disabled communities provide raw material while products are later sold back to them. AI itself is cast as an active agent in four guises: digital accessibility remediator (retroactively 'fixing' inaccessible content rather than preventing it), augmenter of abilities, willing assistant (substituting for human care work in healthcare and education), and enabler of autonomy and independence — the latter relying on ableist assumptions that equate independence with worth. Companies position themselves as global AI ambassadors, moral stewards, and principled leaders through grants, ProcureAccess signings, and 'AI Principles' documents; however, these commitments are voluntary and dynamic (Google removed its categorical bans on weapons and surveillance AI applications in February 2025). Legitimation strategies include co-opting disability-advocacy language ('Nothing About Us Without Us'), downplaying risks with hedged phrases like 'relative accuracy,' and aligning accessibility with market expansion (e.g., citing £30B/year UK economic upside from disabled workforce participation).
Relevance
For accessibility practitioners, this paper is a necessary counterweight to the optimistic press releases that now dominate AI-for-accessibility conversations. It provides a vocabulary — pseudo-participation, technoableism, digital remediator framing, ethics washing — for recognising when vendor claims obscure rather than address structural barriers. Procurement and policy teams can use its findings to ask harder questions: Are disabled people compensated and given authority, or merely testers? Do 'AI accessibility' features prevent inaccessibility upstream or just patch it after the fact? Are ethical commitments enforceable or voluntary? The authors' call for externally enforceable accountability (auditing, the EU Accessibility Act, ProcureAccess) gives organisations a concrete alternative to self-regulation. Limitations: the corpus is U.S.-centric and restricted to for-profit firms, excluding assistive-technology specialists and non-profits whose framings may differ. The paper also does not analyse internal corporate documents, which may show different tensions than public-facing rhetoric.
Tags: artificial intelligence · AI accessibility · critical discourse analysis · disability justice · critical disability studies · technoableism · corporate rhetoric · AI ethics · participatory design · disability representation · ethics washing
Standards referenced: European Accessibility Act