GitHub released a plugin for its Accessibility Scanner that targets alt text that passes automated checks without actually describing an image, according to the GitHub Blog and the plugin's repository.
More than 27% of images on popular websites carry missing or unhelpful alt text, GitHub engineers Taarik Ashenafi and Keenan Zhou wrote, and existing scanners catch only the missing case. Their plugin adds five deterministic rules enabled by default, catching whitespace-only alt text, file names used as descriptions such as "hero.png," placeholder text like "TODO," generic words such as "image" or "logo," and repeated alt text across adjacent images.
A sixth rule is opt-in and uses a vision model with page context, including headings, captions and surrounding text, to judge whether alt text actually describes what is in an image. The team said detecting repeated alt text required checking where images land on the rendered page, not their order in the markup, since visually adjacent images can sit far apart in the underlying code.
The AI rule is off by default, requires an explicit token to enable, redacts URLs and query parameters before sending anything to a model, and sends image bytes only if Azure AI Vision is configured, according to the GitHub Blog. The plugin is available now on GitHub.
Automated accessibility checks have always been good at catching what is missing, not what is useless. A tool built to fail on content quality, not just tag presence, is a template worth borrowing for any team running compliance linting on autopilot.