WCAG 3.0’s March 2026 working draft introduces Bronze-Silver-Gold scoring that compliance tools currently in use aren’t built to handle.
The new standard adds requirements and changes how accessibility gets measured, moving from binary pass/fail checks to graduated scoring systems that evaluate cognitive load, conversation quality, and user control in ways today’s automated tools struggle to assess.
It’s not yet clear how these grades will be enforced or standardized, but it’s going to be good to keep them in mind as we all build and write things moving forward.
The medals system and binary compliance
The March 2026 WCAG 3.0 working draft replaces the familiar A/AA/AAA conformance levels with the aforementioned scoring tiers that measure accessibility quality on graduated scales:
- Bronze represents minimum conformance,
- Silver indicates good accessibility, and
- Gold means excellent user experience.
Your current automated tools (axe-core, WAVE, Lighthouse, etc) are architecturally incapable of making these assessments because they’re built to detect rule violations, not evaluate quality gradients.
Consider a properly labeled form input that meets every WCAG 2.1 technical requirement. Current scanners mark it compliant and move on. Under WCAG 3.0’s draft scoring rubric, that same input might earn Bronze if the label placement creates cognitive load, Silver if it provides clear guidance, or Gold if it anticipates user needs and prevents errors. The technical markup hasn’t changed, but the evaluation has shifted from “does this work?” to “how well does this work?”
The new system uses 0-4 point scales per outcome, requiring tools to assess subjective factors like conversation flow, error recovery effectiveness, and cognitive burden distribution, qualities that demand human judgment or sophisticated AI analysis that goes far beyond current automated pattern matching.
When (not if) AI revolutionizes accessibility, it will need to evaluate these nuanced user experience factors, not just flag missing attributes. Accessibility platforms face substantial technical challenges rebuilding assessment engines for graduated scoring.
AI content generation guidelines
WCAG 3.0’s proposed Outcome 4.2.1 requires AI-generated content to be “appropriate for the intended audience and purpose” (a subjective evaluation). Lighthouse can flag missing alt attributes on AI-generated images, but it cannot detect when those descriptions contain factual errors, cultural insensitivity, or contextually inappropriate language that creates cognitive barriers for users with intellectual disabilities.
This issue becomes clear when AI assistants provide incorrect information: current scanners see properly structured markup and pass the page, while users receive misinformation that violates cognitive accessibility under the new guidelines.
Research comparing automated detection against manual expert evaluation found that AI systems excelled at identifying technical violations but failed completely at assessing content appropriateness, while studies of AI-assisted alt text generation showed that authors could produce descriptive text efficiently, but quality assessment still required human judgment to catch contextual errors that automated tools missed entirely.
The new guidelines explicitly require human evaluation for AI bias detection because automated tools cannot assess whether AI outputs reinforce harmful stereotypes or exclude specific user groups through inappropriate language choices.
More on the cognitive load assessment
More research analyzing LLM effectiveness in detecting WCAG guideline violations yet again found that automated tools excel at identifying technical issues but do the opposite of excel (read: fail miserably) when evaluating semantic accessibility and cognitive load factors.
The new cognitive load assessment criteria require measuring actual user mental effort during AI interactions (impossible without user testing integration that current tools lack). Tenon and axe-core can verify that confidence indicators exist in the markup, but they cannot determine whether a high confidence score creates anxiety for users with anxiety disorders or whether alternative presentation methods would reduce cognitive strain.
The divide between technical compliance and user experience evaluation represents the core challenge facing accessibility vendors as they attempt to retrofit static analysis tools for dynamic AI assessment. Organizations relying on automated scanning (like us) will need entirely new testing methodologies that combine technical validation with cognitive accessibility evaluation.
Tool development lag, and compliance risk
Major accessibility tool vendors estimate substantial adaptation periods before their scanners can handle AI-specific compliance checks. Organizations relying on current automated tools face a dangerous compliance delay where legal requirements outpace technical capabilities. This timeline mismatch creates accumulating technical debt: Every AI feature built to current WCAG 2.1 standards risks Bronze-level failure under 3.0’s scoring system.
The EAA directive enforcement has already accelerated this crisis, with strong regulatory signals pointing toward WCAG 3.0 adoption well before mainstream tools can support the new AI guidelines (likely in 2028 or 2029), though initial court rulings ask more questions than they answer when it comes to enforcement. Organizations following current scanner recommendations build features that satisfy today’s automated checks but create compliance liabilities under emerging standards. Chatbot interfaces that meet WCAG 2.1’s success criteria through proper ARIA labeling might score Bronze under 3.0 if they lack explainable AI decision-making or confidence level indicators.
Organizations that begin manual AI accessibility audits using draft 3.0 criteria gain an advantage over competitors who wait for tool updates. This proactive approach requires more initial investment, but prevents the expensive retrofit cycle that automated-tool-dependent organizations will face when their scanners finally catch up to legal reality. Manual testing protocols enhanced with 3.0 draft checklists consistently outperform any current automated solution for AI accessibility evaluation.
Human-AI hybrid work and the path forward
It feels like the solution, at least for now, is to become one with the machine.
Human-AI hybrid testing platforms need to develop Bronze, Silver, and Gold scoring mechanisms that mirror WCAG 3.0’s structure. Hybrid platforms allow accessibility teams to combine automated scans with human cognitive load assessments to arrive at compliance scores that align with the draft guidelines’ evaluation criteria. Advanced platforms integrate AI-generated explanations with manual review workflows, so teams can test whether machine learning interfaces meet the explainable AI requirements that static scanners can’t currently evaluate.
Custom rule engines offer the most flexible path forward for organizations with technical resources, and we believe every accessibility tool (ours included) needs to start building these as soon as possible.
Specialized AI content testing tools are emerging specifically for WCAG 3.0’s bias detection and transparency requirements. Section 508 guidance indicates that federal agencies are already evaluating dedicated AI accessibility scanners that focus exclusively on machine learning interface compliance rather than attempting to retrofit traditional web testing tools.
Organizations implementing structured human review processes using the latest working draft criteria gain immediate compliance advantages. AI’s impact on digital accessibility requires evaluation methods that combine technical validation with cognitive assessment.
Organizations that master this transition will build genuinely inclusive online experiences that serve all users better, and this should be everyone’s goal.
References
- Akiba, D., Pagliara, S., Truss, M., Nwokoye, C., & Waters, G. (n.d.). AI testing, evaluation, verification and validation for accessibility: a comprehensive framework. Front. Digit. Health, 7, 1679603. https://doi.org/10.3389/fdgth.2025.1679603
- European Accessibility Act (EAA). (2025, December 2). European Commission. https://commission.europa.eu/strategy-and-policy/policies/justice-and-fundamental-rights/disability/european-accessibility-act-eaa_en
- Explainer for W3C Accessibility Guidelines (WCAG) 3.0. (2021, December 7). W3.org. https://www.w3.org/TR/wcag-3.0-explainer/
- He, Z., Huq, S. F., & Malek, S. (2025). Enhancing Web Accessibility: Automated Detection of Issues with Generative AI. Proceedings of the ACM on Software Engineering, 2(FSE), 2264–2287. https://doi.org/10.1145/3729371
- Section508.gov. (2023). Section508.Gov. https://www.section508.gov/develop/incorporating-accessibility-conformance/
- Singh, N., Lucy Lu Wang, & Bragg, J. (2024). FigurA11y: AI Assistance for Writing Scientific Alt Text. https://doi.org/10.1145/3640543.3645212
- W3C Accessibility Guidelines (WCAG) 3.0. (n.d.). Www.w3.org. https://www.w3.org/TR/wcag-3.0/
- W3C Accessibility Guidelines (WCAG) 3.0 publication history. (2024, May 28). W3C. https://www.w3.org/standards/history/wcag-3.0/
