Is Agentic AI Assistive Technology?

You run the digital services team for a local authority. A resident with a motor disability contacts you. They use agentic AI to compare council tax bands, query planning applications, and manage their housing benefit claim. The tool acts on their behalf, the way a support worker might. They’ve been blocked at your website’s bot protection software.

Was your software wrong to block them? The answer is possibly yes. And almost no digital service in the UK is prepared for that possibility.

The analogy that should already be law

Agentic AI is software that performs tasks on behalf of a user. It can browse websites, fill out forms, download documents, draft messages, and complete transactions. For a disabled person, this is more than a convenience. Think of it as the digital equivalent of a support worker or a screen reader.

A Microsoft and EY study published in 2024 tested Microsoft 365 Copilot with over 300 employees who identified as neurodivergent or disabled, across 17 organisations. 91% found it helpful as an assistive technology. Users described it as supporting communication for people with hearing loss, writing for people with dyslexia, and task management for people with dyspraxia.

That study was about a productivity tool but the same logic extends to any task a disabled person delegates to an agent: managing an e-commerce account, applying for a grant, navigating a government portal, booking a medical appointment.

If a support worker can do something on your website on behalf of a disabled client, and agentic AI cannot, you have a contradiction. The Equality Act 2010 may already require you to address that contradiction.

Who is not ready

Government portals, educational platforms, utility services, and financial websites are almost entirely built around the assumption of direct human interaction. They are generally unprepared for the coming age of agent-authorised access: no policy for handling it, and no understanding of the legal complexity it creates.

This is not a criticism of those organisations. It reflects an almost universal gap. The idea that a disabled person might access a public service through agentic AI was speculative even five years ago. It is now a live, practical question. The technology exists and people are using it.

The EAA, which came into force on 28 June 2025, covers consumer banking, e-commerce, and transport services. It requires those services to meet accessibility standards, but it says nothing specific about agentic AI. The disproportionate burden exception further complicates enforcement, allowing providers to argue that full compliance would impose unreasonable costs.

Standards bodies have not yet addressed the gap either. The World Wide Web Consortium (W3C)’s Web Accessibility Initiative has no published guidance on agentic AI. ETSI EN 301 549, the European standard for electronic communications, has no provisions for agentic AI access.

And no UK regulator has published guidance on agentic AI and accessibility. The Equality and Human Rights Commission, the ICO, and Ofcom have all produced work on AI and digital accessibility, but none has addressed this specific question.

The legal thread

Three legal frameworks are most relevant.

The Equality Act 2010 places a duty on service providers to make reasonable adjustments for disabled people. Section 20 requires providers to take positive steps to remove barriers that put disabled people at a substantial disadvantage.

The critical question is whether support for agentic AI counts as a reasonable adjustment. The argument runs like this: if a disabled person uses agentic AI as a form of assistive technology, analogous to a screen reader, voice control software, or a support worker, then a service that blocks that use of agentic AI may be failing to make a reasonable adjustment.

This argument is untested. Equality Act guidance was last updated in 2015, before capable agentic AI existed. No court or tribunal has considered this specific scenario. But the logic is consistent with how the law has been applied to other assistive technology barriers.

The UK Web Accessibility Regulations (SI 2018/952) implement the Web Content Accessibility Guidelines (WCAG) 2.1 Level AA as the standard for public sector digital services. WCAG Principle 4 requires content to be compatible with assistive technologies. If agentic AI is characterised as an assistive technology, blocking it could be argued to violate compatibility requirements.

The EAA requires services to apply universal design principles to systematically remove and prevent barriers. A verification or interaction system that a disabled person’s use of agentic AI cannot navigate may violate that principle. Again: untested, but the direction of travel is clear.

The AI regulation thread

The European Union Artificial Intelligence Act (EU AI Act, Regulation 2024/1689) becomes fully applicable from August 2026. It creates transparency obligations for providers of AI systems that interact with users. Where a service deploys automated decision-making that also interacts with disabled users’ agentic AI, those transparency provisions create additional complexity.

The UK has taken a sector-led, pro-innovation approach to AI regulation rather than passing a single act. Under that approach, existing regulators, including the EHRC, are expected to integrate AI considerations into their work. No specific guidance has emerged yet on the agentic AI and accessibility question.

We expect that as these frameworks bed in, the convergence with existing accessibility obligations will become more apparent.

The precedent we already know about

CAPTCHA is the clearest existing precedent. These are the challenge-response tests designed to distinguish humans from bots. They have been the subject of documented accessibility complaints and enforcement actions for years.

Visual CAPTCHA prevents blind and visually impaired users from accessing protected resources. Screen readers cannot interpret them. Users have been locked out of services as a direct result.

Site owners who deploy CAPTCHA that discriminate against disabled people face potential litigation under Section 508 in the US and comparable laws in other jurisdictions. Multiple frameworks, the Web Accessibility Directive (EU 2016/2102), UK SI 2018/952, and Section 508, create enforceable obligations to provide accessible alternatives.

We see agentic AI as the next escalation of this tension. Instead of blocking a screen reader, a service might block a tool that is doing exactly what a support worker would do on behalf of a disabled person. The mechanism is different but the legal principle feels the same. And the stakes are higher, because agentic AI is capable of far more consequential tasks than any screen reader was designed to perform.

What you can do now

You don’t need a court ruling or a new accessibility guideline to start planning for this. Here is what you can do now.

Audit your bot protection. Does your website’s bot protection system distinguish between automated fraud and legitimate automated access by assistive technology? If it treats all non-human traffic as hostile, it is probably already failing disabled users, not just those using agentic AI.

Review your terms of service. Most services outright prohibit automated access. If a disabled person’s use of agentic AI breaches your terms, you may have a reasonable adjustment obligation to amend them.

Track the regulatory horizon. The EHRC has not published guidance on this question, but it will. The EAA is now in force. The EU AI Act is bedding in. The question of agent-authorised access will not stay in the abstract.

Talk to your users. If disabled people are using agentic AI to access your services, you should know about it. Understanding how people actually use your service is the foundation of accessible design.

Waiting for answers

We can explain the legal and conceptual case. What we are still waiting to see is clear leadership in this area. While there is broader work on AI and accessibility, including the W3C’s AI and Accessibility Research Symposium 2023, which explored the positive and negative impacts of AI in digital accessibility, there is not yet a clearly established, widely adopted position on agentic AI as an assistive technology concern.

Take this as evidence of a gap in the field, not evidence that the problem doesn’t exist.