AI layoffs: the Dossier

This is the record behind our layoffs reporting, chapter by chapter, in the order the dossier keeps it: who is losing work and where, what the companies filed and what they said, who is paying for the machines and for the people, and what the public and the law have made of it.

Every number here carries its label. FACT is a primary record we opened; REPORTED is a tracker's or an outlet's figure, named; CLAIM is a party's own assertion; SIGNAL is what people are saying, counted. Every chart is frozen from its table with its rows, and the same twenty-nine views stand on the Esper Society page and in the instruments console under Society, with their slicers.

The same dossier, read as one argument, is the exposé Counted low.: the AI layoffs, and why every number you have been given is a floor. This page is the dossier itself.

The loudest version of the AI jobs story says the machines are taking work everywhere. The data does not show that. What it shows is narrower and, for the people inside it, worse: the first job is disappearing.

Stanford's Digital Economy Lab follows payroll records for millions of workers. In its August 2026 update, workers aged 22 to 25 in the occupations most exposed to AI sit about 19 percent below where their employment trend would have put them, a gap that has widened in every version of the study. Workers with more experience in the same jobs show no comparable gap. And the same authors write, in plain terms, that they do not see economy-wide displacement. Both things are true at once.

Layoff numbers are counted in thousands. They are lived one household at a time. The aftermath reporting that has been done, most of it in the last two years, tells the same story: months of applications without a reply, contract work at half the pay with no benefits, one income suddenly carrying a mortgage, a return to school in mid-career. The people who spoke to reporters gave their names; we do not repeat them here.

Outplacement firm Challenger, Gray and Christmas counts announced job cuts every month and asks employers why. In the first half of 2026 it counted about 444,000 cuts, and employers attributed about a quarter of them to AI. That is a count of announcements and of what employers say, not of people who lost work because a machine did their job; it is still the closest running tally there is.

Ask Americans, and the verdict is in. In Pew's survey fielded in June 2026, about seven in ten expect there to be fewer jobs in twenty years because of AI. Gallup reports, from May, that nearly eight in ten expect AI to cut jobs within ten years. Pew's global survey across 36 countries finds the same unease almost everywhere.

The data, as chapter 1 shows, has not yet caught up with the fear, except on the bottom rung. That gap between what people believe and what the averages show is itself the story of this moment: it shapes how voters hear politicians, how workers read every layoff announcement, and how quickly a single unverified post can travel. The most-seen claim that AI was doing the work at Oracle reached about 1.5 million views on X and was never confirmed with a source. The most-seen ranked list of 2026's AI layoffs reached over a million and carries a Cisco figure that Cisco's own filing does not disclose.

There are counter-voices with reach of their own. Some argue that many companies were simply overstaffed and that calling a cut "AI" wins an executive credit with investors. On television, the argument is out in the open: on 60 Minutes on 4 October, a former head of AI at two of the largest companies spoke of self-worth and the fracturing of communities, and an MIT economist, Daron Acemoglu, dismissed the framing on air.

The figures. 71% expect fewer US jobs in twenty years (Pew, fielded June 2026, FACT) · 79% expect AI to cut jobs within ten years (Gallup, May 2026, REPORTED) · the Oracle claim about 1.56 million views, confirmed by no reply with a source (SIGNAL and RUMOUR) · the ranked layoff list 1,282,667 views (Kobeissi Letter, read 30 October, SIGNAL)

On 27 August 2026 the US Bureau of Labor Statistics did something it had never done: it rated every occupation it tracks, 831 of them, by how exposed it is to AI, combining the measures published by Microsoft, Anthropic and three academic teams, and set the ratings beside its own ten-year job projections. It is the most authoritative answer there is to the question everyone asks: is my job safe?

The answer cuts against the folklore in two places. First, the most exposed jobs are among the best paid: about a third of American jobs sit in the highest tier, and their median wage is far above that of the lowest tier. This is a story about office work, not factory work. Second, the trades are not untouched. Plumbers, electricians and heating mechanics are rated in the middle, because the paperwork, quoting and diagnosis parts of the job are exposed even if the hands-on work is not. Every underlying measure still ranks them near the bottom, and the middle tier is the one projected to grow fastest.

The programmer is the sharpest case. The Bureau rates software developers as most exposed, and still projects their numbers to grow by about a tenth over the decade; it projects computer programmers, the more routine coding role, to shrink. Customer service representatives and data entry keyers, the jobs that were outsourced a generation ago, are projected to shrink further.

Two things in the public record complicate the safe-harbour story. The trades rescue carries its own rebuttal: the money the largest AI companies put into training electricians buys a four-week course against a five-year, 10,000-hour apprenticeship, and the trades unions' own president called it "a brilliant public relations move"; the viral 1,421 percent rise in Gen Z trade-school enrolment is an artefact of a cohort ageing in from near zero. And retraining, the usual answer for the desk worker, has a weak record: a review of 56 randomised trials found about 1.7 points more employment at about $13,000 a head.

Governments are built on wages. In fiscal 2025, about half of everything the US federal government collected came from individual income tax and another third from payroll taxes: together about five dollars in every six. Corporate tax brought in less than a tenth. Social Security is funded almost entirely by the payroll tax, and its retirement fund is projected to run dry in late 2032; Medicare's hospital fund in 2033.

So the question underneath every AI layoff is not only what happens to the worker, but what happens to the money that funds schools, pensions and hospitals if income moves from wages to profits. By the government's own measures, workers' share of national income is already at or near the lowest since records began, in the late 1940s on one measure and the 1930s on the other. That shows the fall; it does not prove AI caused it. Yale's Budget Lab, modelling AI futures in the current tax system, finds revenue would still rise, but by about half what it would if workers kept their share, with payroll tax revenue falling as inequality grows.

Two answers are on the table. One is a universal income; in the largest US experiment recipients worked a little less, a difference its own lab says is not statistically significant, and every pilot ends (Wales is the one that followed people after the money stopped); paying every adult $12,000 a year would cost about three quarters of federal spending. The other is to tax the machines: a bill in Congress, H.R. 10044, would tax AI usage and AI revenue at rates that rise with unemployment, including a company's own use of AI where it cuts jobs. Neither is law.

Counting what people say on X in a single week is a rough measure, but it shows which fears have taken hold and which have not. In the seven days to 29 October, in English and leaving out reposts, about 4,800 posts talked about AI and layoffs. The phrase "replaced by AI" appeared about 1,600 times. "Learn to code", the advice a generation of workers was given, appeared about 1,100 times, much of it now said with bitterness. Some 900 posts were about new graduates and entry-level work. The most common pairing of all was the trades with AI, about 1,900 posts: the electrician, the plumber and the welder held up as the work a machine cannot take.

Three things stand out. First, the shame and the mockery live mostly in private accounts: the most-liked post on "replaced by AI" in the sample, about 7,000 likes, said some people deserve to be replaced. We do not name or quote private people, so those posts appear here only as numbers. Second, the word "AI-washing", the charge that companies blame AI for cuts made for other reasons, is almost absent, 45 posts in a week, yet the argument behind it is everywhere: hiring slowed before ChatGPT; a buyout is not a layoff; read the numbers before the headline. Third, the fear of the young and the data about the young argue in the same feed: the Washington Post and Entrepreneur on graduates leaving computer science, beside accounts pointing to federal data that show no shock in recent-graduate unemployment yet. That is chapter 1's tension, played out in public.

What is missing is as telling. Searches for the chief executives of the largest AI companies talking about their own layoffs on X found nothing; and none of the three companies whose filed letters say AI changed the work (Block, Cloudflare, Snap) put a layoff statement on X that we could find.

The figures. in the 7 days to 29 October 2026, English, reposts excluded: AI-layoff posts about 4,805; "replaced by AI" 1,569; "learn to code" 1,106; new graduates and entry-level with AI 905; the trades with AI 1,897; fear words with AI and jobs 689; "AI took my job" 262; universal basic income 1,116; shame or stigma words with a layoff 74; "AI-washing" 45 (SIGNAL)

When a large American employer lays people off, federal law (the WARN Act) makes it file a notice with the state, weeks ahead, saying how many people at which site. Those notices are the only place a government counts a layoff by site and by day. Until this year none of them asked why.

That is starting to change, state by state. Since 1 October 2026, Connecticut requires an employer filing a layoff notice to tell the state whether the layoffs relate to its use of AI "or another technological change", a wider question that can be answered truthfully while telling us little; the requirement is one section of a law titled "An Act Concerning Online Safety". New York did not change its law; its labour department changed a form, adding a question inside a login-gated portal whose wording has never been published, and its public dashboard shows no reason, though the export behind it carries one. No official count has been published, so the two tallies in circulation disagree: a law firm reported in May that none of more than 160 notices said yes; the state's own export shows one, from Nespresso for 46 workers in Manhattan, naming "Artificial Intelligence" beside relocation.

The stricter rule may come from New York's legislature. Bill A9581-B has passed both houses and waits only to be delivered to the governor and signed, as the New York Senate's page read on 30 October 2026: every business with 50 or more employees would report each year how many jobs AI cost, including the positions it chose not to fill, which no layoff notice can capture, with a penalty of $500 a day. In Congress, the bipartisan disclosure bill, S.3108, has had one action in its life, referral to committee on 5 November 2025, while its cosponsors kept arriving: two Democrats and two Republicans over twelve months. Its text has no penalty, no threshold and no definition of its test, "substantially due to". The chair of the Senate subcommittee that would hear it said in July that the predicted mass displacement has not begun. Other bills would tax AI where it cuts jobs; none has passed.

The fear is usually told as an American story. The exposure is global, and it runs the other way from most technology shocks: rich countries are more exposed than poor ones, because so much of their work is the desk work AI touches. The International Labour Organization estimates that about a third of jobs in high-income countries are exposed to generative AI, against about a tenth in low-income countries; in rich countries, women's jobs are about twice as exposed as men's, because so many of them are clerical. Where AI could help, tens of millions of workers in poorer countries have no internet connection to use it.