A new Trump-aligned team has taken control of the U.S. Web Design System (USWDS), a federal service used to keep government websites accessible and consistent, and has said it aims to automate as much accessibility testing as possible, according to a September 25 FedScoop report.

The new team removed contributors from the project’s GitHub repository, deployed AI tools, limited or locked discussion threads, and deleted some posts, according to public GitHub activity and sources who spoke to FedScoop. 

In a message titled “our priorities” on the project’s GitHub page, the new USWDS lead said the goal is to automate as much accessibility testing as possible across current and future versions of the system.

The Development Leading to This

USWDS is housed within the General Services Administration (GSA), which helps keep government websites accessible and consistent. The new team, led by Treasury AI engineer Ryan Parker and Treasury CIO Sam Corcos, who has DOGE ties and recently took on additional GSA leadership roles, took control of USWDS in early September.

The Experts Are Concerned

Accessibility experts say current AI tools are not capable of replacing human-led testing, according to FedScoop. Matt Henry, the system’s former engineering lead, said roughly 30% of accessibility testing can be reliably automated. The rest depends on understanding how a real user with accessibility needs actually experiences a site.

Sheri Byrne-Haber, an accessibility strategist who founded VMware’s accessibility program, said automated tools left in complete control tend to miss about half of accessibility violations, in her estimate. Automated testing has always been part of the solution, but was never meant to be the whole solution.

The FedScoop report says their concerns echo a broader trend. A 2026 review of one million top websites by WebAIM found accessibility issues increasing for the first time after years of gradual improvement. The report identified the rise of AI-generated code as a likely contributing factor.

Impact on AI Deployment in Regulated Systems

Accessibility testing depends heavily on understanding how real people, including people using screen readers or other assistive technology, actually experience a website. That is not something current tools can fully assess on their own, according to the experts FedScoop spoke to.

Treating a partially automatable task as fully automatable, without a plan for what the tool reliably misses, points to an AI governance gap.

Then there’s the question of quality control. 

Decisions about what gets automated, and why, had previously unfolded with more public input from the open-source community that relied on the project, and GSA has not explained how it will maintain quality.

For any organization automating a process that requires compliance, whether accessibility, data privacy, or security, automation decisions need documented criteria for what the tool can and can’t reliably catch, and a way for the people affected to see and question that reasoning.