Deque CEO warns AI-generated code accessibility requires human oversight and testing

The reliance on AI-generated code accessibility without rigorous verification transforms a compliance task into a significant operational risk. For streaming and financial platforms, an accessibility failure is functionally equivalent to a service outage, as it renders essential user flows unavailable to specific customer segments. As the European Accessibility Act and other jurisdictions move toward requiring auditable evidence trails, companies must shift from 'compliance checkboxes' to integrated governance. This shift requires a hybrid approach where automation handles volume while human experts focus on complex assistive technology interactions. Watch for a rise in regulatory enforcement actions targeting firms that rely on static vendor reports rather than continuous production monitoring.

Deque Systems has positioned itself at the center of the growing tension between AI-assisted development and accessibility compliance. The company's axe-core testing engine has become the de facto standard for automated accessibility evaluation, and Deque reported in early 2025 that axe-core had been downloaded more than 300 million times, underscoring its dominance in the automated testing space that Preety Kumar argues must complement AI-generated code rather than be replaced by it. The firm has expanded its enterprise offerings to address regulated industries specifically, with Deque Universe serving as a unified platform for accessibility testing across development pipelines. This positioning matters because Kumar's warning targets the exact gap between AI code generation speed and the verification rigor that financial regulators increasingly demand.

Regulatory pressure on digital accessibility is intensifying across multiple jurisdictions, directly raising the stakes for organizations that rely on AI-generated code without adequate testing. The European Accessibility Act entered full enforcement on June 28, 2025, requiring covered digital services to meet WCAG-based standards or face penalties, creating a hard compliance deadline for banks, streaming platforms, and e-commerce operators serving EU customers. In the United States, the Department of Justice issued a final rule in April 2024 that mandates WCAG 2.1 Level AA conformance for state and local government digital services, with compliance deadlines beginning in 2026, signaling that enforcement expectations are expanding beyond the private sector into public-facing digital infrastructure. These regulatory milestones validate Kumar's argument that organizations treating accessibility as a post-hoc checkbox rather than a pipeline-integrated control face escalating legal and financial exposure.

Independent research reinforces the concern that enterprise AI failures persist for many organizations. A 2025 study from the University of Toronto found that large language models produced accessible HTML in fewer than 40 percent of test cases when generating common UI components, with failure rates highest for complex interactive widgets such as date pickers, data tables, and modal dialogs. Separately, WebAIM's annual accessibility analysis of the top one million homepages found that 95.9 percent of pages still contained detectable WCAG failures as of February 2025, a figure that has remained stubbornly flat despite years of increased tooling investment. These benchmarks suggest that even human-written code struggles with accessibility compliance, making the additional layer of AI-generated output without automated verification a compounding risk factor for regulated institutions.