AI and the Jobs Never Posted: Why Young Workers Pay First

The question was put to 1,250 American workers in plain language, with an example attached so nobody could mistake what was being asked. “Since 2023, have you lost a job due to AI, such as AI's automation of your previous job's tasks?”

Roughly ninety-five per cent of them said no.

About three per cent said yes. Between three and four per cent were not sure. The same survey asked whether respondents had landed a job that did not exist before AI, of the AI data labeller or AI compliance specialist variety, and roughly six per cent said they had. It asked whether they had received a promotion or advancement they believed was related to AI, such as their own AI skills, and about nine per cent said yes.

Three times as many workers reported being promoted because of AI as reported being displaced by it. That is the finding, and it landed in late August 2026 into a discourse that had spent four years insisting the opposite was already happening.

The survey was commissioned by Jeffrey C. Dixon, a professor of sociology at the College of the Holy Cross, and fielded by YouGov between 30 July and 4 August 2026. Dixon wrote it up for The Conversation on 27 August, and Fortune republished it two days later under a headline that did the arithmetic for readers who might have missed it: ChatGPT said you would lose your jobs right now, and it is more like three per cent of workers. Dixon, who describes himself in the piece as an AI pessimist, called the result surprising. The study is in progress and has not been peer reviewed. He funded it through a College of the Holy Cross Ruettgers Grand Challenge Award and a professional development account, and he discloses that he also runs a consultancy in responsible AI.

None of which makes the number wrong. It makes it a number, which is a different and more interesting thing than a headline. What follows is an attempt to weigh it properly: to say what three per cent actually means when you multiply it out, what it cannot see, what the strongest evidence on the other side looks like, and why the gap between what the data show and what almost everyone believes is itself the most revealing finding of all.

What Three Per Cent Weighs

Start with the arithmetic the coverage mostly skipped.

There were 163.2 million Americans in employment in September 2026, according to the Bureau of Labor Statistics household survey. Three per cent of that population is roughly 4.9 million people. If you prefer to think of displacement in human rather than percentage terms, the reassuring finding is that somewhere in the region of five million American workers believe artificial intelligence took a job away from them in the past three years. That is not a puddle. It is more than forty times the number of job cuts American employers had publicly attributed to AI in the whole of 2026 up to the end of September.

Then apply the sampling error. Dixon did not publish a margin of error, and at n=1,250 the simple binomial calculation for a proportion of three per cent gives a ninety-five per cent confidence interval of roughly plus or minus one percentage point, so somewhere between about 2.1 and 4.0 per cent. Translated back into people, that is a range of about 3.4 million to 6.5 million. YouGov's panels are weighted rather than randomly drawn, which typically inflates the effective margin further. The honest way to report the finding is therefore not “three per cent” but “somewhere between three and six and a half million workers, probably”.

Now consider the three to four per cent who answered “not sure”. On the job-loss question, the uncertain group is roughly the same size as the affirmative group. If every one of them were in fact displaced by AI and simply did not know it, the true figure would be closer to seven per cent, which would be more than twice the headline. If none of them were, three per cent stands. Nobody can adjudicate this, because the survey asks workers to perform a causal attribution about their own employment that even labour economists with administrative payroll data are reluctant to make.

That is the deepest problem with the instrument, and Dixon says so himself: it is difficult for workers to know for sure whether AI is to blame or to credit for their job changes. A worker learns the reason for their redundancy from their employer, and employers are not neutral narrators. Gallup's Workforce Study, which surveyed 23,717 employed American adults between 4 and 19 February 2026 with a margin of error of plus or minus 0.9 percentage points, found that only one per cent of laid-off workers named AI or automation as the primary reason. The reasons they were given instead were organisational restructuring, cited by fifteen per cent, and budget or cost-cutting and economic conditions, at eleven per cent each. A commercial survey of 1,000 laid-off American workers conducted by ResumeGenius through the Pollfish platform on 29 July 2026 found the opposite pressure at work: fifty-three per cent believed automation or AI had contributed to their job loss, but only twenty-two per cent said their employer had confirmed it. Opt-in commercial panels of this kind are not Gallup, and the finding should be held loosely. But the direction of the gap is instructive. Between one per cent and fifty-three per cent lies an enormous space in which the answer depends entirely on whether you are asking what a worker was told or what a worker suspects.

And then there is the group the survey structurally cannot reach. YouGov sampled employed adults. As Dixon notes in his own caveats, none of the respondents were jobless, whether because of AI or anything else. A survey of people currently in work is a survey of survivors. Anyone pushed out of the labour force entirely between 2023 and 2026 is invisible to it, and participation is still lower than it was at the start of 2025. The American labour force participation rate fell from 62.6 per cent in January 2025 to a low of 61.4 per cent in July 2026, before recovering to 61.6 per cent in August and 61.8 per cent in September. Even after that late-summer recovery, against a civilian noninstitutional population of about 275.6 million, the net fall of 0.8 percentage points since January 2025 is roughly 2.2 million people who no longer count as workers at all. Some retired. Some went back to study. Some gave up. The survey cannot tell you which, because it never spoke to them.

The Canaries Are Still Singing

The strongest evidence against the reassuring reading does not come from a vendor or a pundit. It comes from administrative payroll records covering millions of workers, and it is unambiguous about one specific group.

In August 2026, Erik Brynjolfsson, Bharat Chandar and Ruyu Chen released an updated version of “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence”, drawing on high-frequency ADP payroll data through June 2026. Their first fact agrees with Dixon: they find no evidence of widespread, economy-wide job displacement. Their second does not. Employment of workers aged 22 to 25 in AI-exposed occupations now stands nineteen per cent below where it would be had it kept pace with their less-exposed peers. Experienced workers show no comparable gap. In levels, employment of 22-to-25-year-olds in the two most exposed occupational quintiles fell about eleven per cent between November 2022 and June 2026, while the same age group in the three least-exposed quintiles grew about ten per cent.

Their third fact is that this divergence has widened steadily, from fifteen per cent as of July 2025, when they first documented it, to nineteen per cent by June 2026. Their fifth is that the declines cluster in occupations where AI usage substitutes for human tasks, and that where usage complements workers, employment is flat or rising. Their sixth is that the adjustment is happening through employment rather than base pay, which is what a demand shock looks like when wages are sticky.

The authors are scrupulous about what this is. They call the facts early and descriptive, canaries rather than causal estimates. They note that the patterns attenuate when you control for education, that some divergent trends predate generative AI, and that the effects are more pronounced in the ADP sample than in national survey benchmarks. The Stanford Institute for Economic Policy Research brief published in July 2026 by Neale Mahoney, Erika McEntarfer and Karsen Wahal presses this point harder still. Two papers, it notes, find that hiring in AI-exposed occupations began to decline after the Federal Reserve started raising rates in March 2022, several months before ChatGPT existed. When Brynjolfsson and colleagues added controls for interest-rate exposure and remote work, the entry-level declines did not become notable until 2024.

The same brief is the source of the sentence that has been quoted most often on the other side of the argument. Unemployment among workers in the top quintile of AI exposure has risen by 0.77 percentage points since 2022, while unemployment among the least-exposed workers rose by 0.85 points. The most exposed workers are doing marginally better than everyone else. McEntarfer, who served as Commissioner of the Bureau of Labor Statistics until August 2025, is not a naive reader of labour data.

The Budget Lab at Yale reaches the same place by a different route. In an analysis published on 7 May 2026, Ryan Nunn applied synthetic differences-in-differences to the microdata behind the monthly employment reports, a method chosen precisely because AI-exposed occupations differ systematically from unexposed ones in education, gender composition and, as the Budget Lab found, sensitivity to the business cycle. The estimated impact of AI on employment in the average exposed occupation was close to zero and statistically indistinguishable from it. The same was true for inflation-adjusted hourly wages. Its tracking analysis, updated on 15 September 2026, concludes that the occupational mix is not yet changing in ways that clearly align with AI, and that measures of AI usage show no connection to changes in employment or unemployment.

Both things are true. There is no economy-wide displacement, and there is a nineteen per cent hole where a cohort of young workers should be.

The Job That Is Never Posted

The reason those two findings can coexist is contained in the fourth of Brynjolfsson's six facts, and it is the single most important sentence in this entire literature. The entry-level decline operates primarily through reduced hiring of young workers rather than increased separations.

Nobody was displaced. The job was simply never created.

This is why a survey of workers cannot settle the question, and why the three per cent figure, however carefully measured, is answering a narrower question than the one people think it is answering. A worker who is made redundant knows something happened to them and can be asked about it. A worker who was never hired for a role that a firm quietly decided not to open has nothing to report. There is no exit interview for a job that did not exist. The graduate who sends out two hundred applications and receives four replies does not experience AI displacement. They experience a bad year.

The evidence for hiring avoidance rather than firing is now substantial. Anthropic's own economists, Maxim Massenkoff and Peter McCrory, published research in March 2026 introducing an “observed exposure” measure built from actual Claude usage rather than theoretical capability. They found no systematic increase in unemployment for highly exposed workers since late 2022, and reported that deployment “remains a fraction of what's feasible”, with Claude covering around a third of tasks in the computer and mathematical category against a theoretical ceiling near a hundred per cent. But they also found a fourteen per cent average estimated decline in the job-finding rate for younger workers in exposed occupations, an effect they describe as barely statistically significant. Human resources executives, the SIEPR brief notes, report that AI's effects show up as role consolidation and hiring avoidance rather than as redundancy.

The job postings data tells the same story with unusual clarity. Indeed Hiring Lab found that senior-level postings in the United States rose 14.7 per cent year on year as of May 2026 while entry-level postings fell 7.5 per cent. In software development, the occupation most often held up as the canary, entry-level roles accounted for just 4.5 per cent of postings in the first quarter of 2026 against a senior share of 69.3 per cent. Guillermo Gallacher of the same team found that American software development postings had grown almost fifteen per cent since the launch of Claude Code in late February 2025, while overall postings fell seven per cent. Seventy-one per cent of that increase came from senior roles and thirty-seven per cent from jobs mentioning AI in the title. Postings remained 27.5 per cent below their February 2020 level.

Britain shows the same seniority tilt. Indeed's mid-year UK report, published on 3 August 2026, found graduate job postings down around seven per cent year on year and at their lowest level for the time of year since 2020, with total UK postings down eleven per cent since the start of 2026 and thirty-two per cent below their pre-pandemic baseline, even as AI or related tools appeared in a record 9.4 per cent of UK adverts by the end of June. Britain also raised employer national insurance contributions in April 2025, which is exactly the sort of confound that makes single-cause explanations irresponsible.

A labour market can absorb a technology without a single person being sacked and still leave a generation standing outside the door.

The Layoff That Names Itself

If the displacement is largely invisible, the announcements are anything but. And the announcements are where the story gets its numbers.

Challenger, Gray and Christmas has tracked AI as a distinct reason for job cuts since 2023. Through September 2026, American employers had announced 120,136 job cuts citing AI, approximately twenty-one per cent of all cuts, making it the leading stated reason for the year to date. Through July, the cumulative total since Challenger began tracking the category stood at 184,538. For five consecutive months from March to July, AI led all monthly reasons. In July alone, 10,970 of 33,429 announced cuts named it.

Set that against the denominator. The Bureau of Labor Statistics reported that layoffs and discharges ran at 1.7 million in July and 1.6 million in August, the latest month published on 29 September 2026. Over the nine months to September 2026, that is something like 15 million involuntary separations. The 120,136 AI-attributed announcements amount to well under one per cent of them. Challenger's own year-to-date total of 573,195 announced cuts through September was itself down thirty-nine per cent on the first nine months of 2025, and hiring plans were up three per cent on 2025 year to date. Andy Challenger's summary in the July report was carefully balanced: while AI is shifting the labour market, he said, it is not dismantling it.

What the Challenger data actually measures is worth stating precisely, because it is routinely misread. It measures what companies say when they announce cuts. It is a corpus of corporate communications, not a census of terminated employment. The firm is admirably transparent about how fuzzy the category is. Its July 2026 report walked through the case of Montefiore, a Bronx hospital system that eliminated twelve utilisation review nursing posts after adopting software from Datavant. The New York State Nurses Association said the hospital was replacing human workers with AI software. Hospital leaders called that characterisation misleading. Challenger, unable to establish what the product actually was, filed the cuts under “Technological Update (possibly AI)”, a category that absorbed more than twenty thousand cuts in 2025 alone.

Andy Challenger also said the quiet part clearly. Naming AI in a layoff announcement, he observed, can win over investors while pushing current and prospective employees away, which is why the messaging has swung from hedging to aggressively citing it. And he warned that as regulation takes shape, companies will become more careful in their announcements, which would make tracking AI's real impact on jobs more opaque, not less.

In August 2026 the pattern broke. AI fell to the fourth most-cited reason with 3,462 cuts, its lowest monthly total since December 2025, and restructuring returned to the top of the list with 16,173. September continued the slide: AI fell to fifth with 3,961 cuts, about nine per cent of the month's total, while market and economic conditions led with 8,789. The technology did not become less capable in August or September. The vocabulary changed.

Priors Dressed as Measurements

Underneath all of this sits a measurement problem that is not widely understood outside the research community, and it quietly determines the shape of almost every number quoted so far.

Almost every claim about which jobs are exposed to AI ultimately rests on an exposure index. Those indices increasingly work by asking a large language model to classify occupational tasks. The dominant approach descends from work that formalised a task-level taxonomy of exposure and generated its labels using zero-shot prompts to a single proprietary model, with no retrieved evidence about what AI systems can actually do and no structured chain of reasoning.

In a position paper submitted on 14 May 2026, Luca Mouchel of EPFL and MIT, with Pierre Bouquet and Yossi Sheffi of MIT, argue that this will not do. Their title states the case: jobs' AI exposure should be measured from evidence, not model priors. Their demonstration is more damaging than their argument. Applying zero-shot prompting to all 18,796 occupation-task pairs in O*NET version 30.2, they found that changing nothing but the sampling temperature from 0.0 to 1.0 flipped the assigned exposure label for more than twenty per cent of pairs. One in five judgements about whether AI can do a piece of work was an artefact of a decoding parameter. When the same models were conditioned on retrieved evidence, news articles documenting real deployments and academic abstracts assessing capability, disagreement across temperatures fell below 1.6 per cent, and the grounded labels were preferred by both machine and human evaluators in over seventy-two per cent of disagreement cases while aligning more closely with observed real-world usage.

The authors are careful that this is a position rather than a product. Their point is that when a model's output becomes a policy input, its epistemic status changes. It stops being a prediction and becomes a claim about the world, and claims about the world require evidence, auditing and renewal. Exposure scores, they argue, should meet three standards, reproducibility, external grounding and inspectability, and should be periodically revalidated rather than inherited as immutable ground truth. Given that international bodies including the OECD, the IMF and the International Labour Organization have produced labour-market analyses resting on such measures, this is not an academic quibble. It determines where retraining money goes.

A companion critique arrived six days later. Michelle Yin and Burhan Ogut, in a preprint submitted on 20 May 2026, examine the newer fashion for building exposure measures from AI platform conversation logs. Their observation is elegantly simple: the users in those logs are not the workforce. Holding the method constant and switching platforms changes employment impact estimates by nearly a factor of two, and different distribution channels from the same vendor produce results pointing in opposite directions. After reweighting to the actual employment distribution using BLS demographics, effect sizes fall by between forty-two and ninety-three per cent. Platform users, in other words, look like augmentation. The workforce might not.

Jennifer L. Steele and Isabella Cruz make the same point from a practical angle in a preprint submitted on 16 July 2026, aimed at people trying to choose a career. They compare six recent projections of occupational automation exposure and find marked heterogeneity between them. Their proposed remedy is telling: rather than pick a winner, they average five models, on the reasonable grounds that no single index deserves to be trusted alone. Along the way they cite the figure that ought to anchor every conversation about AI and hiring. In May 2026, Indeed reported that 5.7 per cent of American job postings included AI-related language. That is more than a threefold rise in three years, and it is still barely one advert in seventeen.

Golo Henseke's study of 35 European countries, submitted on 20 April 2026, adds the final complication. Adoption of generative AI ranges from under three per cent to twenty-five per cent depending on country, and while occupational exposure predicts uptake, AI does not, in his phrase, diffuse passively along exposure lines. Skills, task composition and organisational influence matter at the individual level, digitalisation and training investment at the national level. Using a shift-share analysis, he finds no measurable evidence that adoption has yet triggered worker-reported task restructuring. Exposure is not adoption. Adoption is not restructuring. Restructuring is not displacement. Every arrow in that chain is treated as automatic in public argument, and not one of them is.

Whose Apocalypse

Which brings us to the question the data cannot answer. If the evidence has looked like this for two years, why does almost nobody believe it?

Start with the people selling the technology. In May 2025, Dario Amodei, the chief executive of Anthropic, told Axios that AI could eliminate half of all entry-level white-collar jobs within one to five years and push American unemployment to between ten and twenty per cent. It is worth pausing on the structure of that claim. The man forecasting mass displacement was the man selling the displacing machine, and the forecast functioned simultaneously as a warning and as a specification sheet. The SIEPR brief names this dynamic explicitly, observing that fears of an AI jobs apocalypse are often amplified by AI leaders themselves, and citing Amodei by name.

By May 2026 the mood in the vendor community had shifted. Sam Altman told interviewers he did not think there would be the kind of jobs apocalypse that some of the companies in his space talk about, a sentence Eduardo Porter quoted in the Guardian on 25 July 2026 in a piece arguing that the AI jobs apocalypse probably is not coming anytime soon. Three months before that, in February 2026, Altman had gone further, telling an audience that some companies were engaging in AI washing, blaming AI for layoffs they would have carried out anyway.

Which is the second set of incentives. Marc Andreessen put it with unusual candour on the 20VC podcast in March 2026, in remarks reported by Fortune: essentially every large company is overstaffed, he said, by at least twenty-five per cent, most of them by fifty, and a lot of them by seventy-five, and now they all have the silver bullet excuse, which is that it is AI. Martha Gimbel of the Budget Lab made the same point from the other direction in February 2026. If you are a chief executive, she asked, what are you going to say? That you are a really bad chief executive? No. You are going to say the world is changing quickly. Oxford Economics, in research Fortune reported the same month, suspected some firms were trying to dress up layoffs as a good news story rather than a bad one, such as past over-hiring. Forrester, forecasting on 13 January 2026 that six per cent of American jobs, some 10.4 million roles, would be automated by 2030 while twenty per cent would be augmented, simultaneously predicted that more than half of AI-attributed layoffs would be reversed. Its vice president J. P. Gownder was blunt that we may not be heading for an imminent AI job apocalypse.

A redundancy is a confession that management got its headcount wrong. An AI transformation is a strategy. The severance is identical. The share price reaction is not.

Then the media. The most damning inventory of the apocalypse frame appears in the footnotes of the SIEPR brief itself, which supports the phrase “a steady drumbeat of media articles” with citations to Bloomberg on banks laying groundwork for mass workforce cuts, the Wall Street Journal on tens of thousands of white-collar jobs disappearing as AI starts to bite, the New York Times on the AI job apocalypse already being here for some recent graduates, and the Atlantic on something alarming happening to the job market. These are not marginal outlets. They are the institutions that set the terms of elite economic conversation, and their aggregate output has been substantially more confident than the evidence base beneath it.

There is even measurement of the amplification. In a computational analysis submitted on 28 May 2026, Joy Bose compared how messages from technology executives and from laid-off workers circulate on X, examining 763 posts from twenty public accounts. Capital discourse enjoyed a mean amplification advantage of 3.12 times over labour discourse in one study and 4.18 times in another, with a median ratio of 10.77. The asymmetry survived normalisation for follower count, suggesting platform mechanics rather than audience size, and a Reddit replication across 647 posts did not reproduce it. On the platform where technology narratives are set, the executive announcing the transformation is heard several times more loudly than the person it happened to.

And finally the researchers, whose exposure indices, as Mouchel and colleagues showed, can flip one label in five on a temperature setting. When a policymaker reads that a particular occupation is highly exposed, they are usually reading a language model's uninterrogated intuition, laundered through a spreadsheet into the appearance of measurement.

The Arithmetic a Worker Can Actually Use

So what should the worker who has spent four years being told their job is next do with all this?

The first thing is to stop treating “three per cent” and “nineteen per cent” as competing claims. They are measurements of different things. The three per cent is a stock of self-reported displacement among people still in work. The nineteen per cent is a flow of jobs that were never opened to a specific age cohort in a specific set of occupations. You can believe both, and you should, because the best available evidence supports both. What you cannot do is pick the one that suits your prior and call the other hype.

The second is to notice where the risk actually sits. On the current evidence it is not distributed evenly across the workforce, and it is not primarily a risk of being sacked. It is concentrated at the entry point, in occupations where AI substitutes rather than complements, and it takes the form of a door that does not open. The policy response to displacement is severance, retraining and notice periods. The policy response to hiring avoidance is something quite different, and almost nothing in existing employment law addresses it. There is no tribunal claim for the job that was never advertised.

The third is that attribution is a form of power, and it is currently held entirely by one side. The employer knows whether the restructuring was really about AI. The worker can only guess. Challenger's warning that regulatory attention will push firms back towards vaguer language is a warning that this asymmetry is about to deepen. Every statistic in this article that depends on companies volunteering a reason, and that includes the headline Challenger series, is a measurement of corporate messaging strategy wearing the costume of an economic indicator.

The fourth is that the measurement infrastructure is not adequate to the question being asked of it. The Current Population Survey does not ask why. Neither does JOLTS. The most-cited exposure indices are generated by models that will give you a different answer if you nudge a decoding parameter. The most detailed usage data belongs to the companies with the largest commercial interest in what it shows, which is precisely why Anthropic's own economists finding no systematic unemployment increase among exposed workers is more interesting than it would be coming from anyone else. Mouchel and colleagues are right that these numbers now function as claims about the world, and that claims about the world require evidence, auditing and renewal.

The last thing is the simplest. Nine per cent said AI had helped them get promoted, and that is the largest single effect in Dixon's survey. It is also the finding least likely to be quoted, because a story about people acquiring a skill and being rewarded for it does not travel. Dixon himself is properly cautious about it, noting that agentic systems may render today's AI expertise obsolete tomorrow, and that a snapshot cannot tell you whether an advantage will last.

But it is there in the data, three times larger than the displacement figure, and its near-total absence from four years of coverage is not an accident of news judgement. It is what happens when the loudest voices in a debate are the ones selling the machine, the ones explaining a headcount reduction to shareholders, and the ones who have discovered that fear travels further than arithmetic. The apocalypse is a better story than the promotion. That does not make it the true one.

Sources and References

  1. Jeffrey C. Dixon, “I surveyed workers to see if AI had caused job losses and was surprised by the findings”, The Conversation, 27 August 2026. https://theconversation.com/i-surveyed-workers-to-see-if-ai-had-caused-job-losses-and-was-surprised-by-the-findings-290100; republished as “ChatGPT said you'd lose your jobs right now — it's more like 3% of workers”, Fortune, 29 August 2026. https://fortune.com/2026/08/29/ai-workers-survey-job-impact-2026/
  2. Neale Mahoney, Erika McEntarfer and Karsen Wahal, “What is really happening to jobs? Separating AI hype from reality”, Stanford Institute for Economic Policy Research policy brief, July 2026. https://siepr.stanford.edu/publications/policy-brief/what-really-happening-jobs-separating-ai-hype-reality
  3. Eduardo Porter, “The AI jobs apocalypse probably isn't coming anytime soon”, The Guardian, 25 July 2026. https://www.theguardian.com/technology/2026/jul/25/ai-jobs-apocalypse-human-labor
  4. Luca Mouchel, Pierre Bouquet and Yossi Sheffi, “Jobs' AI Exposure Should Be Measured from Evidence, Not Model Priors”, arXiv:2605.15474, 14 May 2026. https://arxiv.org/abs/2605.15474
  5. Golo Henseke, “From Exposure to Adoption: Generative AI in European Workplaces”, arXiv:2604.18849, 20 April 2026. https://arxiv.org/abs/2604.18849
  6. Jennifer L. Steele and Isabella Cruz, “Helping People Choose Careers in the Age of AI”, arXiv:2607.15506, 16 July 2026. https://arxiv.org/abs/2607.15506
  7. Michelle Yin and Burhan Ogut, “Who Uses AI? Platform Selection and the Measurement of Occupational AI Exposure”, arXiv:2605.21743, 20 May 2026. https://arxiv.org/abs/2605.21743
  8. Joy Bose, “Attention Asymmetry in AI Layoff Discourse on X: A Computational Analysis of Capital vs Labour Amplification”, arXiv:2605.29367, 28 May 2026. https://arxiv.org/abs/2605.29367
  9. Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence”, Stanford Digital Economy Lab, August 2026. https://digitaleconomy.stanford.edu/app/uploads/2026/08/Canaries_August2026.pdf
  10. Ryan Nunn, “AI Is Probably Not (Yet) the Reason for Labor Market Weakening”, The Budget Lab at Yale, 7 May 2026. https://budgetlab.yale.edu/research/ai-probably-not-yet-reason-labor-market-weakening
  11. The Budget Lab at Yale, “Tracking the Impact of AI on the Labor Market”, 16 July 2026, updated 15 September 2026. https://budgetlab.yale.edu/research/tracking-impact-ai-labor-market
  12. Sasha Rogelberg, “If AI is roiling the job market, the data isn't showing it, Yale Budget Lab report says, raising questions of 'AI-washing' to justify mass layoffs”, Fortune, 2 February 2026. https://fortune.com/2026/02/02/ai-labor-market-yale-budget-lab-ai-washing/
  13. Challenger, Gray & Christmas, monthly Challenger Reports: “Layoffs Fall, Hiring Picks Up; AI Leads For Fifth Straight Month”, 6 August 2026. https://www.challengergray.com/blog/challenger-report-layoffs-fall-hiring-picks-up-ai-leads-for-fifth-straight-month/; “August Job Cuts Up 58%, Consumer Products, Food Lead”, 2 September 2026. https://www.challengergray.com/blog/challenger-report-august-job-cuts-up-58-consumer-products-food-lead/; “Job Cuts Fall in September; Hiring Plans Up 3% Over 2025 On Weak Early Seasonal Hiring”, 6 October 2026. https://www.challengergray.com/blog/job-cuts-fall-in-september-hiring-plans-up-3-over-2025-on-weak-early-seasonal-hiring/
  14. US Bureau of Labor Statistics, “Job Openings and Labor Turnover Summary”, August 2026 data, released 29 September 2026. https://www.bls.gov/news.release/archives/jolts_09292026.htm; July 2026 data, released 1 September 2026. https://www.bls.gov/news.release/archives/jolts_09012026.htm
  15. US Bureau of Labor Statistics, Labor Force Statistics from the Current Population Survey, series LNS12000000, LNS11300000 and LNU00000000, data through September 2026. https://www.bls.gov/cps/
  16. Mary Page James and Ryan Pendell, “U.S. Workers Continue to Report Downsizing”, Gallup, 17 June 2026. https://www.gallup.com/workplace/711287/workers-continue-report-downsizing.aspx
  17. ResumeGenius, “2026 AI Layoffs Report: Faith In Employers Eroding”, 24 August 2026. https://resumegenius.com/blog/career-advice/ai-layoffs-report
  18. Maxim Massenkoff and Peter McCrory, “Labor Market Impacts of AI: A New Measure and Early Evidence”, Anthropic, March 2026. https://cdn.sanity.io/files/4zrzovbb/website/3f7fd9d552e66269bdb108e207c5d80531d04b8b.pdf
  19. Guillermo Gallacher, “AI and Job Postings: From Destruction to Creation?“, Indeed Hiring Lab, 8 July 2026. https://hiringlab.indeed.com/2026/07/08/ai-and-job-postings-from-destruction-to-creation/
  20. Felix Aidala and Sneha Puri, “The Labor Market Is Tilting Toward Seniority”, Indeed Hiring Lab, 23 July 2026. https://hiringlab.indeed.com/2026/07/23/the-labor-market-is-tilting-toward-seniority/
  21. Indeed Hiring Lab UK, “Indeed's 2026 Mid-Year UK Jobs & Hiring Trends Report: A Labour Market Under Pressure – And in Transition”, 3 August 2026. https://hiringlab.indeed.com/uk/blog/2026/08/03/mid-year-uk-jobs-hiring-trends-report/
  22. Jim VandeHei and Mike Allen, “Behind the Curtain: Top AI CEO foresees white-collar bloodbath”, Axios, 28 May 2025. https://www.axios.com/2025/05/28/ai-jobs-white-collar-unemployment-anthropic
  23. Forrester Research, “Forrester: AI-Led Job Disruption Will Escalate, While Fears Of A Job Apocalypse Are Overstated”, 13 January 2026. https://www.forrester.com/press-newsroom/forrester-impact-ai-jobs-forecast/
  24. Fortune, “Marc Andreessen says AI layoffs are a farce: Companies are 75% overstaffed, and AI is the 'silver bullet excuse' to clean house”, 31 March 2026. https://fortune.com/2026/03/31/marc-andreessen-ai-layoffs-silver-bullet-excuse-overhiring
  25. Sasha Rogelberg, “OpenAI CEO Sam Altman warns 'AI washing' is real”, Fortune, 19 February 2026. https://fortune.com/2026/02/19/sam-altman-confirms-ai-washing-job-displacement-layoffs/

Tim Green

Tim Green UK-based Systems Theorist & Independent Technology Writer

Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.

His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.

ORCID: 0009-0002-0156-9795 Email: tim@smarterarticles.co.uk

Listen to the free weekly SmarterArticles Podcast

Support What We Do

Discuss...