The Admission That Changes the AI Narrative
Speaking virtually at a Commonwealth Bank of Australia conference in Sydney on May 26, OpenAI CEO Sam Altman said he was "delighted to be wrong" about AI's impact on white-collar employment. "I thought there would have been more impact on entry-level white-collar jobs being eliminated by now than has actually happened," Altman told CBA Chief Executive Matt Comyn. "I now think I understand more about why it hasn't." The reversal is significant not merely as a correction of a specific forecast but as a revision of the primary fear narrative that has surrounded AI deployment since ChatGPT's launch in 2022. Altman, whose company is preparing to file for an IPO targeting a $1 trillion valuation, is one of the most influential voices in the AI industry, and his public forecasts — whether bullish or alarming — shape investor expectations, regulatory posture, and public discourse in ways that extend far beyond any single company. His admission that the jobs apocalypse he predicted has not materialised is therefore as consequential for the AI governance debate as any regulatory announcement or earnings report.
The reversal is not Altman's alone. Dario Amodei, CEO of Anthropic, who claimed in 2025 that AI could eliminate 50% of all entry-level white-collar jobs within five years and that unemployment could rise to 10-20%, has also shifted position — now saying automation may actually expand the work people do rather than eliminating it. Goldman Sachs CEO David Solomon, who has argued consistently since late 2025 that the panic was overblown, is now pointing to a century of American economic history to support his view that technology transitions create more work than they displace. The convergence of these reversals from the most prominent voices in AI — the CEO of the company that created ChatGPT, the CEO of its primary competitor, and the CEO of the world's most influential investment bank — is not coincidental. It reflects what the data is actually showing: the Yale Budget Lab, which has been tracking AI's effect on U.S. labour markets since ChatGPT's release, found no meaningful change in occupational mix or unemployment durations through March 2026 for workers in jobs with high AI exposure.
What the Data Actually Shows: A More Complicated Picture
The labour market data through mid-2026 presents a more nuanced picture than either the apocalypse narrative or the complete reversal suggests. Technology sector layoffs through May 2026 have exceeded 144,000, with major reductions at Meta — which announced plans to cut 8,000 jobs as part of its AI restructuring strategy — Cisco, which confirmed approximately 4,000 redundancies with its CEO citing AI as the driver of investment reallocation, and dozens of software and services companies that have reduced headcount in categories where AI tools are directly substituting for human labour. These layoffs are real and represent genuine displacement for the workers affected. They are, however, concentrated in the technology sector's own employment base — the companies building AI are restructuring around it — rather than generating the broad economy-wide unemployment surge that apocalyptic forecasts predicted. The distinction matters: the AI employment impact to date is a within-sector restructuring rather than an economy-wide displacement event, and the affected workers are disproportionately in the technology industry rather than in the entry-level white-collar categories that Altman and Amodei identified as most at risk.
Altman acknowledged that while AI had not eliminated jobs at the scale he feared, he and his executives had been "roughly right" on the technological predictions made when OpenAI launched ChatGPT in 2022. The distinction between being right about the technology and wrong about the social consequences is analytically important. AI systems have achieved capabilities in code generation, document drafting, data analysis, and customer service that match or exceed the forecasts made in 2022. The error was in the model of how those capabilities translate into labour market outcomes — a model that assumed capabilities would be deployed at maximum speed across entire job categories. The actual deployment pattern has been slower, more selective, and more additive than substitutive: companies are using AI to do more with the same headcount rather than doing the same with less headcount, at least in the initial deployment phase. Whether that pattern persists as AI capability advances further and deployment costs continue to fall is the open question that neither Altman's reversal nor the current data fully resolves.
The IPO Timing: Why the Reversal Happened Now
Sam Altman and Dario Amodei are both walking back their AI jobs apocalypse prophecies as they eye blockbuster IPOs. Fortune's framing of the reversals in terms of IPO timing is worth examining carefully. OpenAI is preparing to file its S-1 confidentially with Goldman Sachs and Morgan Stanley advising, targeting a September 2026 listing at a valuation above $1 trillion. Anthropic is planning a public market debut that would make it one of the most valuable AI companies ever listed. Both companies need institutional investors to accept valuations that imply sustained rapid revenue growth, and both know that institutional investors who are pricing their offerings will conduct due diligence on the social and regulatory risk profile of the company — including the probability that AI's labour market impacts generate regulatory responses that constrain the companies' ability to pursue their core business models. A CEO who has publicly forecast that AI will eliminate vast categories of human employment is, in the context of an IPO roadshow, simultaneously arguing that his company should be valued at $1 trillion and that his company's primary product will cause widespread social harm. The cognitive dissonance is commercially inconvenient, and Altman's Sydney appearance resolves it.
The more charitable interpretation — and one that is not incompatible with the IPO timing observation — is that Altman is genuinely updating on evidence. The data from three years of ChatGPT deployment does not support the apocalyptic scenario, and a CEO who is intellectually honest should update his public positions when the evidence diverges from the forecast. The problem is that the same CEO who is now "delighted to be wrong" spent 2024 and 2025 generating significant public anxiety about AI employment impacts with forecasts that he made confidently and repeatedly. The social cost of those forecasts — in terms of worker anxiety, political polarisation around AI regulation, and the policy decisions made on the basis of projected employment impacts — cannot be retrieved simply by announcing at a Sydney conference that the data has not cooperated with the prediction. The reversal is welcome. But it arrives with a credibility cost that will complicate Altman's ability to make the next confident prediction about AI's social trajectory — whatever it turns out to be.
What the Labour Market Evidence Means for AI Policy
The absence of an AI jobs apocalypse through mid-2026 is not evidence that AI will never displace significant employment — it is evidence that the displacement timeline is longer than the most alarming forecasts suggested. The Yale Budget Lab's finding of no meaningful change in unemployment for high-AI-exposure occupations through March 2026 is consistent with an impact that is building gradually rather than arriving catastrophically. The policy implication is significant: governments that structured their AI regulation around imminent mass displacement have more time than they thought to develop the reskilling, social safety net, and labour market adaptation frameworks that structural AI displacement will eventually require. Altman's own company has published a policy document calling for taxes on automated labour and a national public wealth fund partly seeded by AI companies — positions that are remarkably progressive from the CEO of a company preparing to IPO at $1 trillion, and that reflect a genuine, if belated, engagement with the distributional questions that the Vatican's Magnifica Humanitas encyclical, published yesterday, addresses in moral terms. The timing of Altman's reversal — one day after the most significant moral-theological intervention in the AI governance debate — adds another dimension to a week that has been remarkable for the convergence of AI's commercial triumph and its ethical reckoning.