- Who it helps
- Developers and researchers evaluating accessibility testing
- Availability
- October 29 preprint and public implementation; setup needs browser tooling and a model provider. Costs depend on provider usage.
Observe what happens after an action
The study covers ten accessibility criteria and includes expert-labelled cases from real pages. Focus changes and error responses can reveal failures that a static page scan misses.
Keep human review in the loop
The authors report stronger recall on their benchmark. That does not establish coverage of every accessibility requirement or certify a site. No browser audit or repository test was executed here.
What you can do
- Inspect the benchmark criteria before using an automated accessibility report. Open reference ↗
Where the evidence stops
Abstract and README checked November 14; results remain author evidence.
Sources read. Public text checked November 14, 2026 UTC. No installation, model inference or full workflow executed.
Sources & further reading
Paper · paper
InterA11y paper ↗
October 29 abstract read; interaction-based accessibility evaluation is bounded to selected criteria.
Mingyuan Zhong and coauthors · page checked 2026-11-14
GitHub · implementation
InterA11y implementation ↗
README lists Python, Node, browser and provider requirements; no audit run.
jszh · page checked 2026-11-14
Coverage note: Abstract and README checked November 14; results remain author evidence.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
InterA11y combines an AI agent with deterministic browser actions to inspect accessibility failures that require interaction.