GitHub: AI coding can open doors for developers with disabilities, but accessibility isn't automatic

GitHub accessibility head Ed Summers says natural-language development can lower barriers for developers with disabilities. Yet research with Fable found AI models pass automated accessibility checks only 12% of the time by default, rising to 37% with minimal instructions and 60% with a fuller set.

• Baseline accessibility pass rate for AI models was 12%, with the best model at 25%
• Minimal accessibility instructions lifted the aggregate pass rate to 37%, fuller guidance to 60%
• Fable study covered 8 models, 32 prompt cases and 1,280 control samples
• GitHub runs a dedicated accessibility support and feedback channel with triage