Tech Workers Alarmed by AI-Driven Accessibility Testing on Federal Websites
Two government tech officials proposed automating accessibility testing on federal websites, sparking concern among contractors and advocates. They argue that AI cannot replace manual testing required by Section 508, which needs human judgment. The move has drawn immediate backlash from federal agencies and accessibility experts.
The proposed shift centers on the federal web design system, shared code used by at least 600 government websites. Ryan Parker, an AI engineer at Treasury, and Sam Corcos, the department's CIO with ties to DOGE, have signaled that automated tools would handle accessibility checks. The government's own Section 508 guidance explicitly states that manual testing remains necessary, since many accessibility standards depend on context and human judgment.
The GitHub exchange drew sharp criticism from contractors and civic technologists. Laura Gerhardt, who previously implemented the system at multiple agencies, questioned the overhaul's direction and was met with a terse "wrong" from Corcos. Parker repeatedly edited his statements about the scope of automation, and comments raising concerns appeared to be deleted. Employees spoke anonymously, citing fear of retribution.
If AI-driven testing replaces manual review, people with disabilities may find government websites harder to navigate, affecting access to benefits and critical services. Contractors and accessibility experts could lose work or influence, while agencies may face legal exposure under Section 508. The episode also signals how quickly administrative changes can reshape public-facing technology, potentially eroding trust in digital government services among vulnerable populations.