The Web Is Being Made Accessible for AI, Not People (www.techpolicy.press)

Svelte, the popular web framework, has recently updated its documentation to assist artificial intelligences (AI) by introducing a plaintext format for easier navigation by AI systems. This move aligns with a growing trend in tech to make web content more accessible for AI, through structures like the llms.txt convention and other initiatives aimed at enhancing AI's ability to interact with web tools and services. However, this shift raises critical concerns about the implications for accessibility for humans, especially disabled users who historically have not been adequately accommodated. While structured machine-readable formats may benefit AI, they frequently overlook the needs of visually impaired individuals who rely on well-structured HTML and semantic cues to navigate content effectively. The discourse around this development underscores a troubling "ramping automation effect," where accommodations for AI are prioritized over the long-standing accessibility needs of disabled communities. The risk lies in the conflation of "machine-readable" with "accessible," which can lead to superficial improvements that fail to address fundamental human requirements. With accessibility often sidelined until mainstream industries—like AI—demand it, advocates emphasize the need for genuine collaboration between AI developers and disability experts. Ensuring that technology advancements truly serve the public interest necessitates intentional and inclusive design, rather than relying on AI-driven developments as incidental solutions to social justice issues in accessibility.