The layoff blamed on AI, and the data that cannot find it

Companies keep telling workers their jobs were eliminated by AI. The Yale Budget Lab, looking at the entire US labour market through mid-June 2026, cannot find the footprint that story would leave. That mismatch is not a technical curiosity. It is the gap a worker falls through when they try to negotiate severance, claim retraining support, or be taken seriously as someone displaced by a structural shift rather than a management decision.

What the tracker actually says. The Yale Budget Lab updated its labour-market tracker on 15 June. It reports four things: occupational mix is not shifting in ways consistent with AI introduction; AI-usage measures have no statistical link to employment or unemployment changes; a synthetic difference-in-differences design (a quasi-experimental method that builds a statistical control group for AI-exposed occupations) shows no AI-related labour footprint; and the Lab will keep looking.¹ An earlier May analysis from the same team landed in the same place: estimated employment and wage effects for exposed occupations are close to zero and not statistically distinguishable from zero.¹

This is not the same finding as "AI is not affecting work." It is a narrower, more careful claim: through mid-June 2026, the aggregate US labour data does not show a signature that economists can attribute to AI. The Lab is being deliberate about that distinction. The rest of the conversation, mostly, is not.

The corporate story runs on a different track. Through 2025 and into 2026, layoff announcements have routinely cited AI as cause or accelerant. S&P Global's UK survey for 2026 records a net employment-intentions balance of -6 among employers, with AI named as part of the reason.² Consultancies have published headline numbers about tens of millions of jobs disrupted. The Brookings Institution, working the same beat as Yale, calls the current moment the "first inning" of AI's labour effects and stresses that US workers are unevenly equipped to adapt when displacement does come.³

So we have two stories. One is the corporate and consultancy story: AI is reshaping workforces now, and the restructuring you are seeing is the leading edge. The other is the empirical story: at the level the macroeconomic data can see, the reshaping is not yet visible.

Why the gap is doing work. When a firm tells a laid-off worker the role was eliminated because of AI, several things follow from how that claim is received. Severance frameworks differ for "restructuring" versus "technological displacement." Some retraining funds and adjustment programmes are scoped to specific causes. Public sympathy, which shapes the political ceiling on what employers can do, runs hotter for workers displaced by an unstoppable technology than for workers displaced by a management choice to run leaner. The framing is not cosmetic. It changes leverage.

If the population-level data does not support the displacement-by-AI story, then individual firm-level claims need to clear a higher bar. Maybe the firm really did automate a function and cut the role. Maybe the firm cut the role because demand softened, or because it was over-hired in 2021, or because investors wanted margins up, and AI was the available narrative. From the worker's side of the desk, the difference is the difference between being part of a historical transition and being a line item on a quarterly call. The aggregate non-finding is one of the few things workers and their representatives have to push back with.

Two objections worth taking seriously. The first is measurement lag. Agentic tools shipped in late 2025 and early 2026. Quarterly labour surveys take time to register turning points; a footprint that does not exist in June 2026 may exist in June 2027. The Yale Lab is explicit that it is tracking, not concluding. Treating its current update as the end of the argument is a misuse of it.

The second is aggregation. National employment can be flat while specific occupations and specific places churn hard underneath. If AI is destroying customer-service roles in one region and creating prompt-engineering and integration roles in another, net figures can be quiet while individual lives are not. The Brookings adaptive-capacity work is precisely about this: when displacement does land, it will not land evenly, and the workers least able to adapt are the ones who already have the least.³

Both objections are real. Neither rescues the corporate story. A firm citing AI as the cause of a specific layoff is making a claim about its own operations, not about a sectoral churn invisible at the national level. That claim should be checkable. If the firm cannot show which functions were automated, which tools replaced them, and what the before-and-after staffing looked like, the AI attribution is rhetoric.

What the empirical-restraint camp is actually doing. Yale and Brookings are, between them, building a counterweight to the consultancy projection industry. The methodological contrast matters: tracking what has happened in the data versus projecting from capability scores and exposure indices are not the same epistemic activity. The first can be wrong because the data is lagged or mismeasured. The second can be wrong because the model of how capability translates into deployment, and deployment into displacement, is built on assumptions that have not been tested against outcomes. Both are useful. Only one is currently being cited to justify layoffs.

The asymmetry of how these two kinds of evidence get used is the part worth naming. Projected job-loss numbers from consultancy reports get quoted in press releases and earnings calls. The Yale tracker's careful non-finding gets quoted almost nowhere, because it does not serve a transaction. A worker arguing they were not, in fact, displaced by an unstoppable technological wave does not have a McKinsey deck to wave back.

For policymakers and worker representatives, the practical move is to insist on firm-level evidence for firm-level claims. For the rest of us, the move is to stop treating "AI did it" as a self-evident explanation for headcount decisions taken in a quarter when, at the aggregate, AI has not yet done anything the data can see. The displacement may come. When it does, the tracker will show it, and the response we owe workers will need to be larger than severance and sympathy. Until then, the gap between the story and the data is itself a labour issue.

Footnotes

¶