Is AI already causing job losses?

Not across the labour market as a whole, on the best data to the end of summer 2026, and possibly for young workers in the most exposed jobs, where the studies disagree. The UK’s AI minister told a Labour conference meeting on 28 September that there was no evidence so far of an overall reduction in jobs, and that he was preparing a contingency plan anyway. A Federal Reserve governor said the next day that there is “little evidence of significant displacement so far”. A new study of American graduates found no spike in unemployment this summer. The figure of eight million UK jobs is a think tank’s worst-case scenario, described by its own author as avoidable. What an employer decides about its entry-level hiring will show in the data before anything else does.

What the minister said in Liverpool#

Brian Wheeler’s BBC report of 28 September, read here in Finwire’s syndication, has Kanishka Narayan telling a fringe meeting that the government must “take seriously the possibility of an unprecedented impact on the jobs market”, and adding: “If it’s going to hit us in a big way, you are going to want to be as ahead of the curve as possible.” He was careful about the present. UK firms had told him there was “no aggregate impact in the overall labour market today”, though one technology firm had said it hired fewer people because tasks had been automated, “so the actual net impact might not be fully captured”. He named a second gap in the measurement, the “silent adoption” of AI by workers that industry surveys miss, and promised “much better transparency when AI and technology is adopted”. Saskia Koopman at City AM reported that the plan could reach “labour market policy” and “even where we might go on some aspects of regulation”, and set the remarks against the official figures: unemployment at 4.9 per cent in the three months to July, and payrolled employees down 101,000 over the year. Nothing in those figures says why.

Where the eight million comes from#

The number that travelled is not the minister’s. Carsten Jung of the IPPR think tank told the same meeting, on the BBC’s account: “In a worst case scenario you could have about 8 million jobs being negatively affected by this.” He said about 11 per cent of UK jobs are “highly exposed” to replacement by chatbots and that with agentic AI the proportion “could rise to about 60% of tasks across the UK economy”. Three cautions come with it. The first figure counts jobs and the second counts tasks, and they should not be read as one series. Jung conceded that the “stark scenarios” the institute produced in 2024 were “to some extent” wrong. And William Furney at HRreview reported the other end of the same modelling, a scenario with no job losses and a gain worth 13 per cent of GDP, with Jung’s own summary: “Technology isn’t destiny and a jobs apocalypse is not inevitable.” No new IPPR paper carrying these numbers was found; they are spoken remarks about scenarios, and the estate reports them as that.

What the measurements show#

Michael Barr, a Federal Reserve governor, gave the central bank’s reading in a speech in Detroit on 29 September: “While there are some indications that AI may already be a factor limiting new job opportunities for entry-level workers in sectors heavily exposed to AI, across the economy there is little evidence of significant displacement so far.” Robert Fairlie and Jane Wu’s working paper tested the entry-level worry directly on US survey data for June, July and August 2026 and found that graduate “unemployment rates did not spike in summer 2026 relative to summer months in previous years and did not rise in a significant way relative to older college graduates or young workers without a college degree.” Alex Imas and Jacob Schaal’s review of the literature, published the same day as Barr’s speech, sets the disagreement out. On one side is the payroll study that puts employment of 22 to 25 year olds in the most exposed roles 19 per cent below where less-exposed peers would place it. On the other, a paper arguing that the gap tracks remote work, and a Danish study of 25,000 workers that shows, in the reviewers’ words, “precise null effects on earnings, recorded hours, and wages”. Kevin Hassett, the White House economic adviser, said on 28 September that firms using AI see “their employment go up a lot”, Jim Tyson reported at CFO Dive; he named no study.

Why the data may be the last to know#

Both of the minister’s measurement problems are real, and both point the same way. A post that is never advertised produces no redundancy notice and no headline. Joe Rossignol at MacRumors, relaying a Bloomberg report, wrote on 29 September that Apple “recently considered laying off about 5,000 support employees, but it has put those plans on hold indefinitely”, having believed AI agents could take over some of the work. A plan considered and shelved appears in no statistic either. The first instruments are arriving. HRreview reported a government Early Careers Jobs Alliance and research with LinkedIn that found entry-level hiring falling broadly in line with the wider market. California’s SB 951, signed on 30 September, requires notice of mass layoffs “caused in whole or in substantial part by AI or similar systems”, Ryan Golden reported at HR Dive. Until measures like these mature, the estate’s reading of the entry-level evidence on whether AI will replace entry-level jobs stands: the effect, where it exists, runs through hiring that does not happen.

What the argument is really about#

Barr’s speech moves the question from whether to how fast: “If labor market changes happen quickly, it will be hard for workers to adjust and dislocations might be large, whereas a more gradual adoption might permit more orderly adjustments.” He expects substitution in tasks with “clear guardrails and predictable outcomes” and augmentation in work that needs “human judgment, management, coordination and relationships, creativity, or outputs that are hard to measure”, and says now is the time to prepare, “while AI adoption is in its relatively early stages”. Narayan made the companion point at a Resolution Foundation meeting, Kamal Ahmed reported at Fortune: organisations and governments “have agency in whether this ends up being a technology that augments human labor, or is a technology that automates human labor”. That is the estate’s position in a politician’s words. A national job-loss figure is the sum of decisions taken firm by firm about which work a machine does, and the cost that can be measured soonest is what happens to the people who would have learned by doing it, the argument of the missing rungs.

The decision an employer controls#

No employer sets the national number, and every employer sets its own. Rules Before Tools gives the order. Decide which decisions and tasks a machine may take before deciding what that does to headcount. Name who can stop or reverse each one, since several of the firms that cut first later rehired or reversed. Say what people must remain able to do, and before leaving a junior post unfilled write down where the next experienced person comes from, which is the question on how juniors become senior. And keep a record that would let anyone know: posts not opened because of AI, counted as carefully as posts cut. That record is the transparency the minister asked for. It is also the only way a board will learn what its own adoption did. What to tell staff in the meantime is on AI and headcount.

What this does not show#

Nothing here says what happens in 2027. The measured studies are American and Danish; no UK study of AI and employment is cited because none was found this week. The eight million and the 60 per cent are scenario figures spoken at a meeting, and their author says earlier scenarios were partly wrong. The minister’s remarks come from one fringe event as reported by the BBC and three outlets that followed it. A null result cannot rule out a small effect: the Danish study excludes effects above 2 per cent, and the graduate study covers three months and was read in abstract. The 19 per cent gap is disputed and is a distance between two groups, not a fall. The Apple account rests on one Bloomberg report with no company comment in the copy read.