Britain published its own account of not knowing in January. Nine MPs debated the subject in July without one of them mentioning it

THE QUESTION WESTMINSTER SHOULD BE ANSWERING. The Government published an assessment of what AI is doing to British jobs on 28 January. Why has almost nobody in Parliament read it?

WHAT IS IN IT. Assessment of AI capabilities and the impact on the UK labour market runs to five evidence judgements, published by the Department for Science, Innovation and Technology with the AI Security Institute. Its opening paragraph says the available evidence does not yet provide clear answers to many of the questions that matter most for policy.

It is not a document that hides. Around 70 per cent of UK workers are in occupations containing tasks AI could perform or enhance, higher than the United States at around 60, because Britain sells services. Thirty five per cent of the workforce is highly exposed in roles where AI is likely to help. Thirty two per cent is highly exposed in roles where it is likelier to do the work instead.

It reports that UK job adverts in highly exposed occupations fell 38 per cent between 2022 and 2025, against 21 per cent for low-exposure roles. That employment in the digital sector fell in 2024 for the first time in a decade, with the number of 16 to 24 year olds in computer programming down 44 per cent in a single year. And then, in the same passage, that this cannot be attributed to AI, and that hospitality, a sector with low exposure, accounted for 53 per cent of UK job losses between October 2024 and August 2025.

It sets out the case against itself at equal length. Yale's Budget Lab finding the occupational mix is not changing faster than in previous technological transitions. Danish payroll data showing no measurable effect on earnings or hours, even among daily users. It names the two reasons the hiring evidence does not establish cause: exposure is not adoption, and exposure correlates with interest rate sensitivity in exactly the sectors that fell.

WHAT THE MINISTER DID WITH IT. On 8 July, closing a Westminster Hall debate on the societal impact of AI secured by Neil Duncan-Jordan and chaired by Esther McVey, the Parliamentary Secretary at the Cabinet Office, James Frith, told the House the IMF puts around 70 per cent of UK workers in exposed occupations, about half in roles where AI will enhance rather than replace the work. Both figures are in the January assessment. He then said there is little conclusive evidence that AI is reducing employment, which is the assessment's opening sentence in shorter words.

He was quoting his own department's published work, accurately, and he did not say so.

THE GAP. Nine speakers. The member who secured the debate wants an employment levy on firms that replace workers with AI. The opposition front bench went to sovereignty and frontier model safety. Others raised schools, data centres, productivity. Not one of them referred to the document, five months old that day, that had already assembled the evidence they were arguing about without it.

That is not a party failure. It is the whole bench debating a question the state had already published its answer to, which was that it does not know yet and here is precisely what it does not know.

WHAT WE ASKED, AND WHERE WE WERE WRONG. We wrote to the Cabinet Office on 2 September asking what the unit had published, what a labour market assessment recorded in a written answer of 11 March had found, and what monitoring in real time means in practice. We logged the question before we sent it. The department replied on 3 September and pointed to the assessment.

We had not found it. This paper spent Wednesday building an edition on the premise that it did not exist. The editor's note carries that correction in full.

One thing remains genuinely unresolved. We do not know whether the assessment launched in January, referred to in that written answer, is the document published on 28 January or separate work still running. We have asked. We will print the answer.