The FCA spelled out, step by step, how AI takes over. The year before, that sector lost more jobs than any other in Britain
On 6 July 2026 the Financial Conduct Authority published the Mills Review, and inside it is a ladder with five rungs. Operator: you use the machine as a tool. Collaborator: you plan and act with it. Consultant: it weighs the options and recommends, you decide. Approver: it prepares the action, you authorise it. Observer: it acts continuously inside limits you agreed, and you watch what comes out.
That is the regulator's own account of how a person's job changes in financial services between now and 2030. The review is careful. It says few parts of the sector will become fully autonomous, and that not every task travels the whole ladder. It is the first work of its kind commissioned by any regulator anywhere, and it deserves to be read rather than shouted at.
Read it and you find the marketed claim laid out with unusual honesty. AI is offered as the cure for frictions the review names itself: only 9 per cent of consumers use traditional financial advice; just 30 per cent hold life or income protection; around 900,000 people have no bank account; £300bn sits in low-interest accounts. Read those the other way and they describe a market that has been failing those same people for years, with no machines involved at all.
Now the measured reality, from the country's own statisticians.
On 19 March 2026 the ONS published its workforce jobs figures. Financial and insurance activities fell by 78,000 jobs, down 6.6 per cent, from 1,176,000 in December 2024 to 1,098,000 in December 2025. That was the largest annual fall of any of the twenty industry sectors. Roughly one job in fifteen, in a single year.
The caveat travels with the number, and I am not going to bury it. In the same bulletin the ONS discloses that from December 2025 it consolidated how it counts private-sector employee jobs, into an expanded business survey of about 37,000 firms. Response rates and the filling of gaps were affected. Small known differences between newly sampled and existing businesses were amplified in that period. The reference date for the change is the reference date of the fall. Sampling variability on the December 2025 level is plus or minus 37,000, at 95 per cent confidence. The fall is published, accredited and real. Its size sits inside a change of method the statisticians flag themselves. Anyone who uses this figure without that paragraph is using it badly.
Here is the harder discipline. This is not evidence that AI cut these jobs. Nothing above establishes cause, and I have not found anything that does. Vacancies in the sector stood at 33,000 in the three months to February 2026, down 9.2 per cent on the year, close to the 9.5 per cent fall across all sectors. On hiring, finance is not an outlier. Which makes the jobs figure more interesting, not less: it fell further than anywhere else without its hiring collapsing unusually.
Scale, for the denominator. There were 1.17 million financial services jobs in the UK in the first quarter of 2024, about 3.1 per cent of all jobs, the twelfth largest of the twenty sectors by employment and the fourth largest by what it produces.
Parliament got here first, and asked something else. The Treasury Committee reported on AI in financial services on 20 January 2026, a week before the FCA launched its review. Its subject was consumers and the stability of the system. Its chair, Dame Meg Hillier, said on publication that she did not feel confident the UK financial system was prepared for a major AI-related incident, and called that worrying. Note where the worry sits. On incidents. On the system. Not on the people who work in it.
The FCA's engagement paper acknowledges the committee's interest and links to it. The two documents know about each other. The regulator and the committee argue in public about pace, about stress testing, about who belongs inside the rules. On the 1.1 million people, they do not argue. They agree, by not asking. That is not one party's failure. It is what the whole bench has quietly agreed is somebody else's question.
So two things are true, and they are not the same thing. A British regulator has written down, stage by stage, how AI takes over the job. And in the year before it did, that sector shed more jobs than any other in the country. Joining those up is the work. Nobody in authority has taken it on.
The evening spine, dead-centre on the founding question. The marketed claim (autonomy as productivity, AI as the cure for the advice and protection gaps) set against the measured reality (ONS JWS7: the largest annual fall of twenty sectors). The piece refuses the causal claim explicitly and carries the ONS methodology caveat in full rather than in a footnote, because the caveat is the difference between reporting and laundering. The Westminster Gap runs inside the piece, on the whole bench: the regulator and the committee disagree loudly about pace and agree in silence about scope.
Whole bench, on competence. The Treasury Committee (20 Jan 2026) asked about consumers and stability; the FCA (27 Jan 2026) ruled employment out of scope. Each knows about the other; the FCA engagement paper links to the committee. Neither asks what happens to the 1.1 million people who work in the sector. Hillier is used once, paraphrased, attributed, and about preparedness rather than party. No tribe named.
- The Mills Review, FCA. The AI autonomy spectrum: operator, collaborator, consultant, approver, observer (p12, and executive summary p6). Frictions AI is offered as the cure for: advice gap 9%, protection gap 30%, about 900,000 unbanked, £300bn in low-interest accounts (p5). Workforce appears once in the priority recommendations, about the FCA's own staff (p112); employment, incomes and inequality appear in Annex II only as impacts feeding back into demand for financial products (p123).
- ONS, Vacancies and jobs in the UK: March 2026. Workforce jobs by industry (JOBS02, series JWS7): financial and insurance activities down 78,000 (6.6%), 1,176,000 in Dec 2024 to 1,098,000 in Dec 2025, the largest annual fall of twenty sectors. Vacancies by industry (VACS02): 33,000, down 9.2% year on year against an all-sector fall of 9.5%. The same bulletin discloses the December 2025 consolidation of the private-sector employee-jobs collection into an expanded Quarterly Business Survey of about 37,000 businesses, its effect on response rates and imputation, the amplification of known sampling biases, and sampling variability of plus or minus 37,000 on the December 2025 level at 95% confidence.
- House of Commons Library, Financial services in the UK (SN06193). 1.17 million financial services jobs at Q1 2024, 3.1% of all UK jobs; twelfth largest of the twenty sectors by jobs, fourth largest by output.
- FCA Engagement Paper launching the Mills Review. Employment effects placed out of scope by Sheldon Mills in his own introduction, except where relevant to retail markets. The paper acknowledges the Treasury Committee's interest and links to it.
- Treasury Committee statement on its report into AI in financial services (HC 684), quoting the chair, Dame Meg Hillier, on her lack of confidence that the financial system is prepared for a major AI-related incident.