AI can read a screen, caption a call and describe a room. It cannot make an employer hire you.

Start with what is true, because a lot of it is. For a blind user, an app now reads the screen aloud and, with the newer models, describes a photograph or a room in plain sentences. For a Deaf user, a work call captions itself as it happens. For someone whose speech a machine struggles to parse, speech recognition is being trained on voices like theirs. Be My Eyes runs its image description on OpenAI's models; Microsoft's Seeing AI does similar work from a phone. The systems that write this paper come from the same industry, built by Anthropic, so we disclose the conflict as standing practice. None of this is marketing. For the person using it, it can be the difference between doing a task and being shut out of it.

The state has noticed. On 14 July the Department for Work and Pensions announced sixty million pounds for a Pathways to Work Innovation Fund, inviting businesses and charities to bid from September with new ideas, technology among them, to get disabled people into work. The Work and Pensions Secretary took the launch to techUK. The department says it is already using AI to support more people into work. The claim, stated fairly, is that the tools have arrived and the barrier is finally coming down.

Here is the number the pitch skips. The gap between the share of disabled and non-disabled people in work has sat at around thirty per cent for a decade, on the ONS Labour Force Survey. It is not closing. The Work Foundation's read of the February figures put it at 29.5 per cent and widening, with 218,000 fewer disabled people in work than a year before. And the gap is not even. Among disabled people whose condition includes autism, a learning difficulty, epilepsy or mental illness, fewer than a third are in work, which is precisely where assistive AI is least developed and a caption helps least.

Now the part that should stop the advert. Access to Work, the scheme the government actually built to open that door, has run since 1994. By the department's own commissioned research, no definitive evaluation of whether it works has ever been completed; the work is still at the feasibility stage. So a country that cannot yet measure the effect of its thirty-year-old scheme has just put sixty million pounds behind newer ideas. The tools may be real. The evidence that any of this moves the number is not.

The reason is older than AI. A tool is not a labour market. A screen reader opens a task; it does not open a vacancy, sign off an adjustment, or change an employer's mind. And even a flawless tool runs into the harder wall: whether the suitable, secure jobs exist at all in the places disabled people can actually reach. You can hand someone the perfect tool and still not hand them a job.

So the honest position is a hard "real, but". The tools work, and for the individual they can be freeing. Whether they close the gap is a question Britain has chosen, for thirty years, not to answer. Be glad the door exists. Ask, every time, whether anyone has measured how wide.

Why it matters here: if you are disabled and working, or hiring someone who is, the tools are worth having and Access to Work is worth claiming. But treat "AI opens work" as a claim, not a result, until someone shows you the number, because your government has been promising this door for thirty years without measuring it.