'40% fear AI will take their job': what the data actually says

A data analyst I know spent a Sunday evening reading her company’s quarterly memo three times. It mentioned, in one careful sentence, that the firm was “leaning into AI-driven efficiency across functions.” No layoff notice. No manager hinting at one. Just that sentence, a feed full of tech-cut headlines, and a quiet certainty that she should probably update her resume tonight. By Monday she had convinced herself her role had maybe eighteen months left.

That feeling — not a pink slip, but the dread of one — is now the median experience. Mercer’s Global Talent Trends report for 2026, which surveyed around 12,000 workers and leaders, found that 40 percent of employees fear losing their job to AI. Two years earlier the same fear sat at 28 percent. In the space of two annual surveys, anxiety that used to belong to the worried minority became close to the room’s default mood.

For people building data, ML, and AI careers, this is a strange place to stand: we are the ones deploying the models everyone else is nervous about, and we are also nervous. So it is worth separating the fear from the evidence — because the evidence is more mixed, and frankly more useful, than the headline number suggests.

The fear is real, and so are some of the cuts

Take the anxiety seriously rather than waving it away, because part of it is grounded.

AI-attributed layoffs in the United States reached roughly 55,000 in 2025 — about twelve times the 2023 figure, with the vast majority, around 51,000 of those jobs, landing in tech. If you work in software or data, the cuts were not happening to some distant industry. They were happening in yours, to people whose LinkedIn profiles look a lot like your own. That proximity is exactly why the fear spreads faster among technical workers than the raw numbers alone would predict.

The leadership side reinforces it. Mercer found that essentially every CEO it asked — on the order of 99 percent — expects AI to drive at least some headcount reduction within two years. Read that carefully: it does not say companies will halve their workforce, only that almost no executive expects AI to cause zero job change. But “some reduction, somewhere, within two years” is a sentence that, when it reaches employees, tends to lose its qualifiers and arrive as a threat.

The human cost shows up in the wellbeing data. Over the same window Mercer tracked, the share of workers saying they “feel good at work” fell from 66 percent to 44 percent. You do not need a regression to connect those dots: a workforce that suspects its own tools are auditioning for its jobs is not one that feels good clocking in.

Now put the number in its frame

Here is the part the headlines tend to drop. Those 55,000 AI-attributed layoffs happened inside a year of roughly 1.1 million total US job cuts. AI’s share was under 5 percent.

That ratio matters for how you read the moment. The dominant causes of layoffs in 2025 were the ordinary ones — cost-cutting after over-hiring, weak demand, restructuring, interest rates, the long hangover from the 2021–22 hiring boom. AI was a real and growing slice, but it was a slice, not the pie. When a single sentence about “AI efficiency” makes you feel your role is doomed, the base rate is quietly arguing for more calm than the feed is giving you.

There is a second, sharper complication: AI is often the story a company tells, not the reason it cut. “We’re restructuring around AI” sounds like strategy; “we over-hired and demand softened” sounds like a mistake. Given the choice, plenty of leaders reach for the first framing even when the second is closer to the truth. The cleanest evidence that the AI-cut story is partly theater comes from asking whether the cuts actually worked.

The cuts often don’t pay off

If companies were laying people off because AI genuinely did their work, you would expect those companies to be doing better afterward. Largely, they are not.

Gartner found that among firms that piloted AI, about 80 percent reported some workforce reduction — but there was no correlation between those cuts and higher return on investment. Let that land. The companies that trimmed staff in AI’s name were not, as a group, more profitable for having done so. The layoff and the payoff came uncoupled.

The disappointment runs deeper than headcount. Only about 27 percent of CEOs said their AI investment had actually met expectations — meaning roughly three in four, by their own admission, are not getting what they hoped from the technology they are reorganizing around. For a data professional, this is the most important slide in the deck: the enterprise AI rollout is, at this stage, far more uneven and oversold than the press releases imply. The gap between “we deployed AI” and “AI delivered” is wide, and a lot of careers are being decided inside that gap on vibes rather than results.

None of this means the layoffs were painless for the people in them. It means the strategic logic offered for them frequently does not survive contact with the numbers — which is a reason to distrust the inevitability the layoffs seem to broadcast, not a reason to dismiss the disruption.

The macro forecast is a net gain, not a cliff

Zoom out from the quarter to the decade and the picture inverts. The World Economic Forum’s Future of Jobs report for 2025 projects that by 2030, AI and broader automation will displace about 92 million jobs globally and create about 170 million — a net gain of roughly 78 million.

A few honest caveats before you exhale. A net gain is not a personal guarantee — the displaced 92 million and the created 170 million are not the same people, skills, or places, and “a job exists somewhere for someone” is cold comfort if the one that vanished was yours. Forecasts this far out are directional, built on what employers say they expect, and employers are not famous for predicting their own behavior. Treat the +78 million as a description of churn, not a promise of safety.

But churn is genuinely different from collapse. Automation history is mostly the work changing shape, not disappearing — spreadsheets did not end accounting, they ended manual ledgers and created financial analysis. The WEF numbers suggest AI is the same kind of event: violent at the level of specific tasks, roughly neutral-to-positive for total employment, and brutal precisely at the seams where old roles end and new ones have not yet formed.

Where the disruption is sharpest: the entry rung

If there is one place the optimism should narrow, it is the bottom of the ladder — and this is the part our readers should not breeze past.

The tasks AI does best right now are, awkwardly, the ones junior roles used to be built from: first-draft code, boilerplate, data cleaning, routine analysis, the “go figure out this small well-scoped thing.” Mercer flagged workers roughly 22 to 27 years old as facing the highest displacement risk for exactly this reason — automate the simple tasks and you hollow out the apprenticeship those tasks used to be. A Stanford analysis put a number on the early tremor: employment among software developers aged 22 to 25 fell by nearly 20 percent between 2022 and 2025, the stretch in which AI coding tools went mainstream.

This should change what you do, not just how you feel. The risk for an early-career data or ML person is less “a model will do your whole job” and more “the rungs you would have climbed to build judgment are being sawn off, so you have to find another way up.” The work that survives and compounds is what AI cannot yet do alone: framing the ambiguous problem, deciding which question is worth answering, validating output you cannot fully trust, owning the consequences when the model is confidently wrong. Lean there. The roughly three in four CEOs disappointed by AI’s results are, in effect, describing a job for the humans who can close that gap.

The quiet aggravator: nobody explains the AI

There is one finding I keep coming back to, because companies could fix it tomorrow and mostly don’t. Gallup found that 44 percent of employees say AI is already in use at their workplace — but only 22 percent say leadership has explained how it will be applied. That is a two-to-one transparency gap, and into the silence rushes exactly the kind of Sunday-night spiral my analyst friend fell into.

This is the lever, and it is almost insultingly cheap. The anxiety is driven not only by the technology but by watching it arrive without a word about what it means for you. A manager who says plainly, “here is what we are automating, here is what we are not, here is what we want you doing instead,” removes more dread than any reassurance about job security could, because it replaces an unknown with a plan. If you lead even a small team, that conversation is the highest-return thing you can do this quarter. If you are an individual contributor, you are entitled to ask for it — and the answer, or the inability to give one, tells you a great deal.

So what do you actually do with this

Hold two facts at once, because both are true and neither cancels the other. The cuts are real, the entry rung is genuinely thinning, and your worry is not irrational. And those cuts are a small share of total layoffs, the companies making them mostly aren’t winning, and the decade-scale forecast is more jobs, not fewer. Panic reads only the first half; denial reads only the second. The clear-eyed move reads both, then acts on the part within your control.

Concretely, for an AI-era data or ML career: assume routine, well-scoped work will keep automating, and deliberately spend your hours where judgment, framing, and accountability live — the exact places that 27-percent-satisfaction figure says machines are still failing. Get close enough to the tools to know what they do well and where they confidently lie, so you are the person who can tell the difference in a room. And read your own employer’s silence as a data point: a company using AI heavily but explaining it to nobody is not necessarily about to cut you, but it is telling you something about how it will handle the transition.

My analyst friend still has her job. She also has a much better answer than dread — she asked her manager the blunt question, learned the team was using AI to kill the reporting grind she hated anyway, and spent the freed time on the analysis work that actually gets her promoted. The number that scared her was real. It just wasn’t the whole story. Almost none of the scary numbers are.