Sam Altman, who spent much of 2025 warning that AI would gut entry-level white-collar work, now says he got it wrong. Speaking by video link to a Commonwealth Bank of Australia conference in Sydney in late May, the OpenAI chief executive said the labor market damage he expected simply has not shown up.
"I'm delighted to be wrong about this," he told CBA chief executive Matt Comyn, adding that he had expected more entry-level elimination "by now than has actually happened." Executives are rarely this cheerful about being incorrect, and the timing invites a raised eyebrow: OpenAI is preparing for an IPO, and "we are not going to destroy your job" is a friendlier message to carry into a roadshow than the one he was delivering a year ago.
What changed, according to him
Altman's explanation is that the "human part" of work turned out to be stickier than he modeled. People want to deal with people. He offered himself as evidence, saying he had once handed his Slack and email over to AI and has since gone back to answering some of it himself. Which is an anecdote, not data, and he presented it as one.
He said something sharper to CNBC in early June: the companies loudest about AI-driven layoffs are the ones adopting AI least. "The companies that I know that have adopted AI the most are also the ones hiring the most," he said. That is a convenient correlation for a man selling AI, and "the companies I know" is not a sample frame. Still, the broader evidence points the same direction. Independent researchers have found that the visible strain sits at the bottom rung, in junior hiring that quietly never opens, rather than in layoff waves.
The policy contradiction nobody is resolving
Here is the part that does not fit neatly. Weeks before Altman started saying the apocalypse was cancelled, OpenAI published a policy document premised on the opposite. "Industrial Policy for the Intelligence Age: Ideas to Keep People First" runs 13 pages and proposes taxes on automated labor, a national public wealth fund seeded partly by AI companies, and a pilot for a 32-hour work week paid at 40 hours.
Those are not the proposals of a company that thinks the labor market is fine. They are proposals for a company that expects the tax base to erode because payroll shrinks. The document says as much, warning that AI-driven growth could hollow out the revenue that funds Social Security and unemployment insurance.
So which is it? OpenAI's answer, roughly, is that the near term is calm and the superintelligence-era disruption is still coming. Maybe. It is also true that a robot tax proposal reads very differently depending on whether you think anyone will ever collect it.
Who is left holding the doom
Almost nobody, apparently. Anthropic's Dario Amodei, who in 2025 predicted AI could erase half of entry-level white-collar jobs and push unemployment to 10 or 20 percent, has also softened. Peter Wildeford of the AI Policy Network offered a blunter theory for the collective softening: public opinion. Americans have turned negative on AI, and the industry has responded with an about-face.
That reading is uncharitable. It is also hard to dismiss, given how synchronized the reversals were and how close they land to two IPOs.
None of the OpenAI proposals are law. No bill has been introduced. Altman called the document a starting point rather than a prescription, and the company closed its feedback inbox on June 9 after more than 400 submissions, with grant recipients still under review.
