OpenAI releases Explainable Contrast Checker: AI that cites rules, not just scores

AI · 5 min read

The Explainable Contrast Checker inspects design files or screenshots and produces human-readable rationales tied to specific WCAG success criteria, including recommended token swaps (e.g., change "accent-500" to "accent-700") and suggested role/label updates. Unlike black-box tools, it generates step-by-step reasoning and links to standards and test cases that developers can validate.

Developers can integrate the module into CI and design token pipelines: when a proposed token change might affect brand color harmony, the checker offers alternatives that balance contrast with brand intent. It also outputs machine-readable patches that can update token files in Git and open pull requests with changelogs describing accessibility rationale.

OpenAI stresses that the module is a decision-support tool and recommends human review; the company published a responsible-use guide and a set of evaluation datasets that measure whether recommendations are actionable and minimally disruptive to design systems. Early feedback from agencies shows it reduces low-level contrast fixes and accelerates consensus around token choices.