Artificial intelligence (AI) is increasingly embedded in digital products, from customer service chatbots and voice assistants to automated decision-making tools. However, these systems can create barriers for people with disabilities when accessibility, fairness, and user choice are overlooked.
Canada’s CAN-ASC-6.2:2025 - Accessible and Equitable Artificial Intelligence Systems provides a framework for addressing these challenges. Published by Accessibility Standards Canada in December 2025, it focuses on making AI accessible to people with disabilities, preventing unfair outcomes, and embedding accessibility and equity throughout the AI lifecycle.
For digital product teams, the standard offers a practical basis for reviewing AI-powered interfaces, system outputs, development tools, testing processes, and governance practices.
What is Canada’s Accessible AI Standard?
CAN-ASC-6.2:2025 is Canada's first National Standard focused specifically on accessible and equitable AI. It considers whether people with disabilities can participate in developing, evaluating, procuring, managing, and using AI systems.
The standard addresses the following three areas:
- Accessible AI: Ensuring people with disabilities can use AI systems and participate in their development and management.
- Equitable AI: Ensuring AI systems provide fair benefits and do not unfairly exclude or disadvantage people with disabilities.
- Accessible AI education and training: Helping people understand AI, its effects, and the choices available to them.
The standard is voluntary and can be used in whole or in part for conformity assessment. It should not be confused with a standalone legal mandate applying identically to every Canadian organization. Teams must separately assess the laws, regulations, and other accessibility requirements applicable to their products and services.
What should be reviewed in digital assets?
The standard is most useful when translated into concrete product reviews. Teams should assess not only whether an AI interface is accessible, but also how the system is developed, what it produces, whom it serves, and how users can challenge its outcomes.
Review the accessibility of AI interfaces
An AI feature can introduce barriers even when the surrounding website or application is accessible. Chatbots, voice assistants, recommendation tools, and automated forms must support users with different sensory, physical, cognitive, and communication needs.
- Keyboard navigation: Can users access, operate, and exit AI features without a mouse?
- Screen reader compatibility: Are prompts, responses, loading states, errors, and dynamically updated content announced appropriately?
- Speech and voice controls: Can users interact through alternatives when speech recognition or voice output is unsuitable?
- Cognitive accessibility: Are instructions clear, responses understandable, and interactions predictable?
- User control: Can users pause, repeat, edit, or correct interactions where appropriate?
Check whether AI-generated outputs are accessible
Accessible input does not guarantee accessible output. AI systems may generate content that excludes users, even when the interface itself meets technical accessibility requirements.
For example, an AI assistant might generate an image without meaningful alternative text, a document with an incorrect heading structure, or captions that misrepresent spoken content.
Product teams should review:
- Text and document: Are headings, lists, links, tables, and reading order structured correctly?
- Images and visual content: Do descriptions convey relevant information, and are essential details missing?
- Audio and video: Are captions, transcripts, and other relevant alternatives available and accurate?
- Language and readability: Is the content understandable for its intended audience?
- Generated code and interfaces: Do AI-generated components preserve semantic markup, keyboard access, and compatibility with assistive technology?
Testing should include representative outputs rather than relying on a single successful example.
Explore free accessibility checkers and tools to evaluate digital assets accessibility.
Evaluate fairness across disability groups
AI systems may perform differently for people with disabilities because of gaps in training data, assumptions built into model design, or inadequate testing. These differences can affect access to services, employment opportunities, information, and other important activities.
Teams should review:
- Does the training or evaluation data adequately represent people with disabilities?
- Have outputs and decisions been tested for disability-related disparities?
- Could the system disadvantage people who use assistive technologies, alternative communication methods, or nonstandard speech patterns?
- Is the evaluation method appropriate for the intended use and the potential impact of errors?
- Can the team identify, investigate, and correct unfair outcomes?
The standard emphasizes equitable access to AI’s benefits and the need to prevent unfair harm. Teams should therefore assess real-world outcomes, not simply overall model accuracy.
Preserve meaningful user choice and human support
AI should not become the only route to an essential service when that creates barriers for people with disabilities. Users may need another interaction method, assistance from a person, or an opportunity to challenge an AI-assisted decision.
Digital product teams should verify that:
- Users receive clear information about an AI system’s purpose and role in decisions.
- Accessible feedback and complaint channels are available.
- Users can request human assistance or an appropriate alternative when needed.
- There is a process for reviewing and correcting harmful or inaccurate outcomes.
Consider a virtual assistant used for customer support. If it cannot understand a user’s communication method, the user should not be trapped in a repetitive automated conversation. An accessible escalation route can help them reach an appropriate human representative.
These safeguards reflect the standard’s emphasis on autonomy, accessible feedback, and the ability to challenge decisions.
Make accessibility part of the AI development lifecycle
Accessibility must extend beyond the final product. Development, deployment, monitoring, and maintenance tools also need to be accessible so people with disabilities can participate in AI-related work.
Product teams should review the tools and processes used to:
- Design, code, and configure AI systems.
- Create and evaluate training datasets.
- Test model outputs and user experiences.
- Deploy, monitor, and maintain AI features.
- Report accessibility issues and track remediation.
Accessibility should be included in product requirements, design reviews, development workflows, quality assurance, and release approvals.
Establish ongoing monitoring and accountability
AI systems change as models, datasets, prompts, integrations, and user needs evolve.
Organizations should establish clear ownership for accessibility and equity throughout the AI lifecycle.
A practical monitoring process should include:
- Defined accessibility and equity objectives for each AI feature.
- Periodic evaluations of output quality and disability-related disparities.
- Clear responsibilities for investigating incidents and implementing fixes.
- Regression testing after model, interface, or data changes.
Teams should also retain evidence of decisions, tests, user feedback, and remediation. This creates a clearer basis for tracking improvements and explaining how accessibility risks are managed.
Involve people with disabilities in design and testing
Accessibility cannot be fully evaluated from technical specifications alone. People with disabilities bring firsthand experience of barriers that development teams may overlook.
CAN-ASC-6.2:2025 emphasizes participation throughout the AI lifecycle, including design, development, procurement, evaluation, governance, and monitoring.
Digital asset teams should:
- Include people with disabilities in research and usability testing from early design stages.
- Evaluate how users complete essential tasks, such as navigating an AI chatbot.
- Test with screen readers, magnification, keyboard navigation, voice input, and other relevant assistive technologies.
- Recruit participants with varied disabilities and communication needs.
- Provide accessible ways for users to share feedback and report problems.
- Incorporate findings into product decisions and verify that fixes resolve the reported barriers.
Testing should reflect the product’s actual context. For example, a conversational AI tool may need different evaluation scenarios from an AI-powered platform or document-generation service.
How does Canada’s AI standard relate to existing accessibility requirements?
CAN-ASC-6.2:2025 complements established ICT accessibility standards (CAN/ASC-EN 301 549:2024) rather than replacing them. Product teams should use these relevant technical standards and applicable legal requirements.
An AI interface may meet conventional technical accessibility requirements while its outputs, decision-making processes, or user support still create inequitable outcomes. Conversely, a fair AI model can remain unusable if its interface is inaccessible.
A complete review should therefore consider technical accessibility, equitable outcomes, user autonomy, and organizational accountability together.
Turn Accessible AI Principles into Action
Canada’s Accessible AI Standard reinforces an essential principle: AI innovation must not come at the cost of accessibility, equity, or user autonomy. Digital product teams should evaluate AI interfaces, generated content, disability-related bias, and human oversight throughout the AI lifecycle.
Reviewing digital assets considering Canada’s accessible AI standard? Start by identifying accessibility barriers in your website content, forms, and interactive experiences. Our website accessibility audit services combine automated checks with manual testing for WCAG issues and provide actionable remediation recommendations. Request an audit to prioritize improvements and create more accessible digital experiences.
