Accessibility of AI
Can disabled and neurodivergent people use the AI system?
This includes accessible interfaces, keyboard and speech interaction, screen-reader compatibility, multimodal alternatives and manageable conversational experiences.
AI accessibility is fundamentally cognitive accessibility
AI accessibility
Artificial intelligence can remove accessibility barriers. It can also create entirely new ones.
Of AImake AI systems accessible
Through AIuse AI to reduce barriers
Under AIprotect people affected by AI
The COGAI definition
AI must not become the barrier
AI accessibility concerns whether disabled and neurodivergent people can access AI, understand its outputs, obtain equitable outcomes and remain in control when AI influences important parts of their lives.
It is broader than using AI to generate captions or alternative text. It includes the accessibility of an AI product, the barriers AI can remove, and the consequences people face when AI makes, supports or mediates decisions.
A complete definition
Three areas of AI accessibility
An organisation needs to consider all three. Improving one does not cancel out harm in another.
Accessibility through AI
Can AI remove or reduce barriers?
AI can support captioning, alternative text, summarisation, explanation, adaptation, communication and other assistive experiences—when its output is reliable and checked.
Accessibility under AI
What happens when AI influences a person?
People need to understand, question and correct AI-mediated decisions. The system must not interpret disability, neurodivergence, communication or behaviour unfairly.
From interface to institution
Four connected practice strands
Accessible AI systems
Interfaces, conversational systems, multimodal AI, agents, assistive technology compatibility, keyboard and speech interaction, and WCAG.
Cognitive accessibility of AI
Prompting burden, comprehension, memory demand, uncertainty, hallucination, information overload, explainability, error recovery, trust calibration and human agency.
AI equality and risk
Recruitment, financial services, education, employment, healthcare, automated decisions, disability bias, neurodivergence, vulnerable users and high-risk journeys.
AI accessibility governance
Procurement, design requirements, assurance, testing, evidence, organisational ownership, risk escalation and ongoing monitoring.
COGAI AI Accessibility Framework
Six capabilities every AI interaction should protect
This human-centred sequence can guide discovery, design reviews, procurement, testing, assurance and incident learning. It is an assessment lens, not a substitute for applicable standards or legal advice.
Access
Can I perceive, reach and operate it in a way that works for me?
Understand
Can I understand the interaction, output and language without unnecessary effort?
Evaluate
Can I recognise uncertainty, limitations, evidence and possible error?
Decide
Can I make an informed choice without manipulation or avoidable pressure?
Control
Can I correct, challenge, opt out or reach a person where appropriate?
Recover
Can I identify what went wrong and put it right without disproportionate harm?
Turn the framework into practice
Build specialist AI accessibility capability
Start with how AI systems communicate and behave, then develop the skills to design AI-supported decisions around human cognition, uncertainty and control.
CAIS
Cognitive Accessibility in AI Systems
Learn where AI can break human understanding and how to create clearer, more structured, interpretable and governable AI experiences.
Explore the CAIS courseCDD
Cognitive Decision Design for AI Systems
Learn to design AI-supported decisions around attention, memory, trust and uncertainty while preserving agency, challenge and recovery.
Explore the CDD courseAccessibility under AI
The interface is only the beginning
Risk becomes especially important when AI influences access to work, learning, money, healthcare, support or essential services.
- Recruitment and employment
- Education and assessment
- Financial services and insurance
- Healthcare and care
- Customer-service agents
- Workplace copilots and autonomous agents
Governance and evidence
Make accessibility part of the AI lifecycle
AI accessibility should be owned across accessibility, product, design, engineering, procurement, legal, risk, human resources, data and service operations. Teams need requirements before purchase or build, participation by disabled people, evidence-based testing, routes for challenge and correction, and monitoring after deployment.
The EU AI Act addresses accessibility, universal design, vulnerable groups and AI practices that exploit vulnerabilities related to disability. Its requirements vary by system and application date, so organisations should assess their specific context rather than rely on a general summary.
Read the EU AI Act on EUR-Lex · See the European Commission AI Act overview · Explore W3C cognitive accessibility guidance
Clear answers
Common questions about AI accessibility
What is AI accessibility?
AI accessibility is the practice of ensuring disabled and neurodivergent people can access and understand AI, benefit from it, and remain protected and in control when AI influences decisions or services.
Is AI accessibility just using AI to create alt text and captions?
No. Those are examples of accessibility through AI. AI accessibility also covers whether the AI itself is usable and whether AI-mediated decisions, recommendations and interactions are understandable, fair and challengeable.
How is AI accessibility related to cognitive accessibility?
AI interactions can place heavy demands on attention, memory, language, judgement and trust. Cognitive accessibility asks whether people can understand what the AI is doing, assess its output, make an informed decision and recover from error.
Who owns AI accessibility in an organisation?
It is a shared responsibility across accessibility, AI governance, product, design, engineering, procurement, legal, risk, data, human resources and service operations.
The COGAI position
People must remain able to understand, decide, control and recover.
AI accessibility is not solely a technical feature or an accessibility-team task. It is an organisational capability.