The AI jobs apocalypse is being called off, and some of the people doing the calling are the ones who predicted it.

That framing comes from The Decoder, which has Sam Altman and Dario Amodei walking back their job-apocalypse predictions, and from a Forbes column of 25 July by contributor Don Muir. There is real material behind it. Muir notes that Ford's Jim Farley told the Aspen Ideas Festival last year AI would replace "literally half of all white-collar workers in the U.S." while his own company had rehired 350 people — veteran engineers in Muir’s telling; Bloomberg’s headline called them quality inspectors — a figure Bloomberg reported as accumulating over three years, and which we covered in June: Ford brought experienced staff back precisely to make the automation work. Muir quotes Jeff Bezos telling VivaTech that "AI is going to create a labor shortage… because it's going to make it possible for people to identify more problems." And on the figures Muir cites from EY-Parthenon's CEO Outlook survey, the share of chief executives expecting AI investment to cut headcount significantly fell from roughly 46% in January 2025 to 20% in May.

Altman's own contribution, reported on 12 July, is worth reading in full rather than in summary:

so far at least, i'm pretty sure AI has been net job-creating. this was not what i expected--although i was much less pessimistic than others, i thought by this level of capability we'd have seen some impact. it is possible this direction keeps going!

That is a man reporting a surprise and hedging it three ways — "so far at least", "pretty sure", "it is possible". It is not a retraction, and it does not claim to know the future.

What the research actually says

The evidence everyone reaches for is the Budget Lab at Yale, which tracks this monthly. Its current edition is dated 15 June 2026 and leads with four key takeaways. Three carry the findings:

The occupational mix is not yet changing in ways that clearly align with the introduction of AI into the workforce.

Measures of AI usage show no connection to changes in employment or unemployment.

A synthetic differences-in-differences analysis of AI exposure does not yet clearly indicate an AI-related labor market footprint.

The fourth is a commitment to keep updating the analysis. Two of the four are tensed — "not yet changing", and "does not yet clearly indicate" — and neither says the effect is absent; they say it has not shown up in these instruments. The middle takeaway is flatter than either, and is the strongest line here for the reassuring reading.

There is no prose to go with them. The June edition publishes those takeaways alongside an embedded interactive tracker; the narrative report the page used to carry was dropped between 28 May and 15 June. The Lab's document endpoint still returns the 16 April 2026 edition, covering the March 2026 CPS, and that remains the most recent full write-up available — so the quotations that follow are the April edition's, and are labelled as such. It put its conclusion this way:

Currently, measures of exposure, automation, and augmentation show no sign of being related to changes in employment or unemployment.

With no substantial acceleration detected in how fast the job mix is changing, the authors wrote, "there is nothing meaningful we can either attribute or misattribute to AI." Note "no substantial": the April edition's own first takeaway says the occupational mix is changing more quickly than in the past, but that "it is not a large difference and predates the widespread introduction of AI in the workforce."

And the report was emphatic about what it does not establish, in language the coverage tends to drop:

Of course, our analysis is not predictive of the future.

Its summary:

While anxiety over the effects of AI on today's labor market is widespread, our data suggests it remains largely speculative. The picture of AI's impact on the labor market that emerges from our data is one that largely reflects stability, not major disruption at an economy-wide level. While generative AI looks likely to join the ranks of transformative, general-purpose technologies, it is too soon to tell how disruptive the technology will be to jobs.

"Too soon to tell" is not "it isn't happening." This is a measurement exercise reporting that its instruments have not moved, and saying so carefully. It is being read as a verdict.

Two details survive only if you read the report itself. The scope is explicit — stability "at an economy-wide level" — and the authors note that exposure "does not equate to automation or job loss." The finding is about aggregates. It says nothing about whether any particular person lost work to AI.

The second cuts against the reassuring reading. On Anthropic's usage data for February 2026 and November 2025, the April edition records that "both of these samples indicate that observed usage is more likely to be associated with automation than augmentation." The measure most often invoked to argue that AI assists rather than replaces reads, on both of those samples, the other way. The April edition does not present that as a shift — both readings point the same direction — but it is not the augmentation story either.

The one signal moving is an uptick in how differently recent college graduates aged 20–24 are distributed across occupations compared with those aged 25–34. It "remains at the high end of the historical range," and the authors caution twice about reading it, writing that results "should be interpreted with caution particularly given small sample sizes." A thread to watch, not a finding to bank.

The part the walk-back story gets wrong

Forbes is careful here, and deserves the credit: Muir writes only that in his June essay Amodei "clarified that he issued his original warnings so policymakers and companies could adapt, rather than to sow fear." That is accurate.

The stronger claim — that Amodei now treats automation as a productivity multiplier rather than a job killer — is The Decoder's, and it is sourced rather than invented. The remarks behind it were made at an Anthropic press briefing in Lower Manhattan on 5 May, where he appeared alongside JPMorgan chief executive Jamie Dimon, and were reported by Fortune the same day: "If you automate 90% of the job, then everyone does the 10% of the job. And the 10% kind of expands to be 100% of what people do and kind of 10xs their productivity."

The Decoder's 27 May report of that was fair, and could not have anticipated what came next. The problem is the 12 July item, which repeats the characterisation in the present tense — Amodei is "now calling automation a productivity multiplier rather than a job killer" — a month after he published an essay saying something rather different, and without mentioning it.

That essay is Policy on the AI Exponential, published on 10 June, and what it proposes is a displacement programme.

He is direct about motive, and this much of the framing is fair:

I have warned about job displacement in interviews and essays because I want both policymakers and the private sector to have the best chance to adapt and respond, not because I am trying to be a "prophet of doom".

Clarifying why you issued a warning is not withdrawing it. In the same essay he writes that "enduring job displacement is undesirable and dangerous, and we should do everything we can to minimize or prevent it, not to bring it about," and that "policy can be most helpful in buying us time to do that work, by slowing down job loss and providing economically for those likely to be affected." The proposals include expanding government statistics "to more carefully track AI job displacement"; wage insurance, retention tax incentives to discourage layoffs and training grants, to "slow or reduce job displacement"; and this:

If AI-driven labor displacement ends up being large in magnitude and permanently drives down the demand for labor, it will likely be necessary to go beyond mere incentive programs to long-term income support for a significant fraction of the labor force. Mechanisms such as universal basic income could be financed through taxes on relevant companies or raising the capital gains tax.

Alongside it, Anthropic released a policy framework for job displacement, for which, the essay says, "we intend to provide substantial financial backing."

A chief executive contemplating tax-funded universal basic income as a contingency, and putting company money behind a displacement framework, has not concluded the problem went away. Both things are true at once: he has pushed back on being cast as a doom-monger, and he still expects displacement serious enough to plan for. The later and more substantial of those two positions is the one that travelled least.

What to take from it

The honest summary is narrower than either version on offer. Nearly four years on from ChatGPT's launch, the aggregate US labour market has not visibly bent, on the best public measurement available. The forecasters have adjusted, and Altman's own wording carries its limits. The researchers whose work is cited as the all-clear write "not yet" twice in four takeaways and say it is too soon to tell. And the executive most associated with the alarm spent June proposing wage insurance and a universal basic income contingency.

One caution about corroboration. The Brookings commentary often cited alongside the Yale work shares an author with it: Molly Kinder is a Budget Lab co-author and a Brookings Metro senior fellow. Those are not two independent confirmations; they are one research lineage reported twice.

None of which means the apocalypse is coming. It means the strongest available evidence supports a duller claim than either camp is selling: nothing measurable yet, economy-wide, with the instruments we have — and the people closest to the data saying plainly that this is not a forecast.

The aggregate also has nothing to say about any particular employer. For the other side of that, see Oracle's 21,000 job cuts, which the company blamed on AI in writing — and, for where the labour market is visibly repricing rather than shrinking, AI skills now paying a 62% premium.