Language selection

Government of Canada / Gouvernement du Canada

Search

Backgrounder 3 – Artificial intelligence (AI), accessibility and standards

Note: This backgrounder applies only to Phase 2 of the Digital Technology Accessibility Regulations.

On this page

Introduction

The Phase 1 Digital Technology Accessibility Regulations were finalized in December 2025. They focus on the accessibility of websites, web applications, mobile applications and digital documents. Phase 2 will cover additional areas, including artificial intelligence (AI).

AI can be complex. This backgrounder provides a plain-language overview of AI and possible accessibility barriers to help participants take part in the discussion. It supports early discussions with the disability community, Indigenous partners, industry associations and regulated organizations on the second phase of the Digital Technology Accessibility Regulations under the Accessible Canada Act (ACA).

Canadians already interact with AI in many parts of everyday life. For example, AI can answer customer service questions, summarize information, support decisions, or create written and visual content.

AI can also improve accessibility. For example, it can provide automatic captions, improve image descriptions and offer support based on a person's needs. However, AI can also create new accessibility barriers for persons with disabilities.

This backgrounder also introduces CAN/ASC 6.2:2025, Accessible and Equitable Artificial Intelligence Systems. It is intended to support discussion and does not set requirements.

What is AI

Artificial intelligence (AI) refers to machine-based systems that use data to generate predictions, content or decisions with varying degrees of autonomy, helping perform tasks or influence outcomes. This description is adapted from the definition of an AI system in the European Union Artificial Intelligence Act (European Parliament & Council of the European Union, 2024).

Many AI systems identify patterns in data and use those patterns to predict outcomes, sort information into categories, or create content such as text or images.

As AI becomes more common in websites, applications, internal processes, and automated service tools, it is becoming an important part of the digital technology landscape.

Not all AI systems present the same level of risk. Some applications, such as summarizing information or generating draft content, may have relatively limited impacts if errors occur. Other applications, such as systems used to support decisions related to employment, access to programs and services, education, healthcare, or public safety, may have more significant consequences for individuals. Appropriate governance, oversight, and risk management measures may therefore vary depending on how AI is used.

Accessibility barriers associated with AI systems

AI can create unique challenges and accessibility barriers for persons with disabilities in different ways. The examples below include some potential barriers as well as factors that can increase the likelihood of accessibility problems.

1. Failure of AI-enabled systems to recognize a range of abilities

AI-enabled systems are often designed and trained using data that reflects common patterns of communication, movement, behaviour, or interaction. As a result, they may not adequately recognize or accommodate the wide range of abilities and disability experiences found in the population.

Some AI systems rely on characteristics such as a person's face, voice, or movement to identify them or interpret their behaviour. If the data used to develop these systems do not adequately represent differences in how people look, speak, move, or interact, the systems may not work as well for persons with disabilities.

For example, speech-recognition systems may have difficulty understanding individuals with speech disabilities or atypical speech patterns. Facial-recognition systems may not work reliably for individuals with certain physical disabilities or facial differences.

These limitations can create barriers to accessing services, completing transactions, verifying identity, participating in educational or workplace activities, or interacting with digital technologies. In some cases, individuals may not be aware that an AI system has failed to recognize or accommodate their abilities, which can make the resulting barrier more difficult to identify or address.

Considering a broad range of disability experiences during the design, development, procurement, testing, and ongoing evaluation of AI systems can help reduce these risks and support more accessible and inclusive outcomes.

2. Inaccurate AI inferences about disability, needs, or abilities

Some AI systems may make assumptions about a person's characteristics, needs, preferences, or abilities based on patterns in how they interact with digital services. These assumptions may be made even when disability information has not been collected directly.

For example, a system may make conclusions about a person based on how quickly they type or respond when using a digital service.

When these assumptions are inaccurate, they can influence the information, services, or recommendations a person receives. For example, an AI system may recommend services based on what it assumes the person needs or prefers, rather than what the person wants. This could limit a person's choices or make it harder for them to access information or services that meet their actual needs.

Ensuring that AI systems are designed, monitored, and evaluated for fairness can help reduce the risk of unintended bias and inaccurate assumptions.

3. Biased or unfair outcomes in AI-supported decision-making

AI can be used to support decisions about people, such as when organizations hire employees or determine who is eligible for a program or service.

AI systems can produce unfair results when the data used to develop them do not accurately represent persons with disabilities. These systems may learn patterns from historical data that reflect existing biases and reproduce them when making or supporting new decisions.

For example, some employers use AI tools to monitor how people work and how productive they are. If these tools do not account for the different ways persons with disabilities may work, they could incorrectly assess a person with a disability as being less productive.

4. Privacy and security risks involving disability data

Some AI systems collect, process, infer, or generate information that may reveal whether a person has a disability or provide information about their health, functioning, behaviour, or support needs.

Disability-related information may be sensitive personal information. If collected, used, disclosed, or retained inappropriately, it could affect an individual's privacy and potentially contribute to unfair treatment, discrimination, or loss of trust.

In some cases, individuals may not be aware that disability-related information is being inferred based on how they interact with a digital service, rather than information they have provided directly.

Organizations should consider privacy, security, transparency, and consent when designing and deploying AI systems that process disability-related information.

5. Limited transparency about the use of AI

People may not always know when AI has been used to make or support a decision about them. They may also not receive a clear explanation of how AI contributed to the decision. This can make it difficult to understand the outcome, ask questions, or challenge the decision.

It may also be unclear who is responsible for decisions involving AI systems. Clear accountability mechanisms can help individuals understand who is responsible for an AI-supported decision.

6. Limited access to human oversight and support in AI-enabled systems

Some AI systems make decisions or recommend actions without providing a clear option for human review.

Without an opportunity to seek clarification, provide additional information, or request human review, people may be unable to correct an inaccurate result or explain circumstances that the system did not consider.

For example, an AI system may recommend that a person is not eligible for a service based on information in their application. If the system does not account for disability-related circumstances, access to human support may allow the person to provide relevant information and have the recommendation reviewed by a qualified person.

7. Barriers created by AI chatbots and voice tools

AI chatbots, virtual assistants, and voice-based systems can provide convenient ways to access information and services. However, they may create accessibility barriers when they are not designed to accommodate a wide range of users and disability experiences. For example, some AI tools may:

These barriers can make it harder for some persons with disabilities to access information or services through these tools.

When AI chatbots and voice assistants are used in customer service and public-facing services, considering accessibility throughout their design, testing, and deployment can help prevent or reduce these barriers.

8. AI-generated content may not conform to accessibility standards

AI can generate documents, webpages, presentations, and other content. However, the content it generates may not follow accessibility standards or established accessibility practices.

For example, AI-generated content may use headings incorrectly, rely on colour alone to communicate information, use unclear link text, create inaccessible tables, or use formatting that does not support assistive technologies.

These issues can make content difficult or impossible for some people with disabilities to navigate, understand, or use. AI-generated content may therefore need to be reviewed and corrected to ensure that it meets applicable accessibility requirements.

9. Inaccurate AI-generated information used for accessibility

AI can be used to provide information that improves accessibility, such as captions, image descriptions, or directions. When these supports work well, they can make information and services easier to access.

However, AI-generated information is not always accurate or reliable. Incorrect captions, descriptions, or directions can mislead people and, in some cases, create safety risks.

For example, an AI-generated image description may omit important details that a person needs to understand the image.

Information can be accessible in format without being accurate in content. Both accessibility and accuracy are important to ensuring meaningful access. Human review can help identify errors before people rely on AI-generated information.

10. Barriers to AI education and literacy

As AI is used at workplaces, in public services, and in everyday life, learning how to use it is becoming more important.

Some persons with disabilities may face barriers when AI training materials, educational resources, or learning platforms are not accessible. These barriers can make it more difficult to develop skills that can support participation in education, employment, civic life, and daily activities.

Accessible AI education and literacy can help support the meaningful participation of persons with disabilities as AI is used across different areas of society.

Cumulative impact of AI accessibility barriers

An accessibility barrier associated with an AI system may seem minor when it happens once. However, when the same barrier happens repeatedly, its effects can build up over time and create significant challenges for persons with disabilities.

For example, repeatedly having trouble verifying your identity could make it harder to access a service. Over time, these repeated barriers may also become harder to recognize and address.

The cumulative effects of accessibility barriers may be particularly significant when individuals interact with AI systems frequently across different areas of daily life, including employment, education, healthcare, financial services, transportation, and public services.

Because these impacts can develop gradually, they may be difficult for organizations to identify through isolated complaints or individual transactions.

Regular monitoring, user feedback, and ongoing engagement with persons with disabilities can help organizations identify and address accessibility barriers before their effects become more widespread.

Nothing without us: involving persons with disabilities in AI development

AI systems are more likely to meet the needs of persons with disabilities when they are involved in how the systems are planned, developed, tested, and used.

Persons with disabilities bring lived experience and expertise that can help identify barriers, accessibility needs and problems that might otherwise be missed.

Meaningful involvement is also important when organizations procure AI systems. Persons with disabilities can help identify accessibility needs, inform requirements, and identify potential barriers before an AI system is acquired or deployed.

Meaningful engagement can help organizations understand different disability experiences, make AI systems more accessible, and build trust in how AI is used.

How AI barriers can affect people differently

The accessibility barriers described in this document may affect people differently. Persons with disabilities are not a homogeneous group, and experiences with AI systems can vary significantly depending on a range of factors.

A person's disability can intersect with other aspects of their identity and circumstances, such as their age, race, gender, language, income or Indigeneity. These factors can affect how they experience AI systems.

AI systems may not work as well for everyone if the data used to develop them do not reflect a wide range of people and experiences. This can increase the risk of incorrect or unfair outcomes for some groups.

For example, a voice-based AI system may work less accurately for a person with a disability who speaks with an accent that was not well represented in the data used to develop the system.

Considering a wide range of disability experiences can help identify barriers that might otherwise be missed. This supports the development of more accessible, inclusive, and equitable AI systems.

AI governance frameworks

A growing number of domestic and international frameworks, standards, and policies are being developed to support the responsible design and use of AI systems. These initiatives address issues such as transparency, accountability, risk management, human oversight, accessibility, and the protection of rights.

These frameworks differ in scope and purpose, but they share a common goal of helping organizations identify and address potential risks while supporting responsible and inclusive uses of AI.

Summary of Accessible Standards Canada's AI standard

CAN/ASC 6.2:2025, Accessible and Equitable Artificial Intelligence Systems, was published by Accessibility Standards Canada in December 2025. It is intended to help organizations create and use AI systems that are accessible, inclusive and fair. Following the standard is voluntary unless it is adopted into regulation.

The standard has four main sections on AI accessibility and equity:

Sources and further reading

This backgrounder draws on research, guidance, and standards from organizations that are frequently referenced in discussions about accessibility, disability rights, and artificial intelligence, including:

References

European Parliament & Council of the European Union. (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union.

Page details

2026-10-19