Sam Altman Walks Back His Own AI Jobs Apocalypse

Sam Altman told a Commonwealth Bank of Australia conference in Sydney on May 26 that the AI boom is not going to produce a "jobs apocalypse" and that he and OpenAI had been "pretty wrong" about the social and economic forecasts they made around ChatGPT's launch in 2022. He pointed out that entry-level white-collar work has held up better than he expected, and that a "human part" of most jobs has turned out to be stubbornly difficult to automate (Euronews, May 26).

This is a notable climbdown. Altman has spent the better part of three years giving interviews in which he warned that whole job categories were about to vanish and that universal basic income was the only humane endpoint. He is now closer to the line that the actual data has supported all along, which is that the labour market has wobbled but not collapsed.

Amodei is softening too

Anthropic's Dario Amodei made the most-quoted version of the apocalypse claim in 2025, when he told Axios that AI could erase about half of all entry-level white-collar jobs within five years. In the past fortnight he has shifted to a more cautious register, talking about which roles will change rather than which will disappear. Fortune has a useful side-by-side of the old and new framings.

The timing is hard to ignore. OpenAI confidentially filed for an IPO at up to a trillion-dollar valuation last week, and Anthropic is closing a $30 billion round at a $900 billion valuation. Both companies need a public narrative in which AI is transformative without being so socially destructive that regulators feel obliged to intervene. The earlier "we are going to eliminate everyone's job" line was useful when the goal was to make models sound powerful. It is much less useful when the goal is to reassure retail investors and lawmakers that the technology is safe to scale.

What the data actually shows

US Bureau of Labor Statistics figures on software developer employment, often cited as the canary in the coal mine, show the workforce slightly larger in early 2026 than in early 2024. Job postings are well below their 2021 peak, but they were already falling before ChatGPT existed, mainly as part of the post-pandemic tech correction. Pieces from Yale's School of Management and the Society for Human Resource Management have argued that what is being labelled as "AI layoffs" is often a polite cover for over-hiring during the 2021 boom and ordinary cost-cutting since.

This site covered the same pattern earlier this week: tech-industry layoffs have been blamed on AI more loudly than the data actually supports. None of which means the long-term picture is benign. The honest version is that AI is changing how work gets done at a faster rate than past technologies, and slower than the loudest predictions of 2023. The executives now saying the apocalypse was overstated are the same executives who first sold it. Worth remembering when the forecast shifts again.