Three AI CEOs Walked Back Their Job Apocalypse Predictions. Finally.

Key Takeaways:
– Sam Altman told a Sydney banking conference on May 26, 2026, that his own job-apocalypse prediction was “pretty wrong”. Just a year after hammering that same warning.
– Dario Amodei walked back his 2025 claim that AI could wipe out half of white-collar jobs, reframing automation as a pure productivity play instead.
– U.S. civilian employment is up 145% since 1962. Every single automation wave. Data-center construction alone added 200,000 roles since 2022.
– Yale Budget Lab found zero measurable shift in occupational mix or unemployment duration for high AI-exposure jobs since ChatGPT dropped in late 2022.
– Goldman Sachs CEO David Solomon and Deutsche Bank’s Torsten Slok both say the same thing: automation is real, mass job loss isn’t here yet.

—

Here’s what happened in the last week of May 2026.

Three of the most consequential people in AI essentially said: we were wrong. Not in a blog post nobody reads. Not in a quiet podcast.

At public events, on camera, with their titles attached.

Sam Altman told Commonwealth Bank of Australia CEO Matt Comyn at a May 26 banking conference in Sydney that he expected far more entry-level white-collar destruction by now.

That hadn’t happened. His exact word: “pretty wrong.”

Dario Amodei, whose Anthropic is quietly prepping an IPO alongside OpenAI, reversed his 2025 claim that AI could eliminate half of all white-collar roles.

The messaging shift was abrupt.

Jensen Huang told Channel NewsAsia that CEOs blaming AI for layoffs were just being lazy. That’s a hell of a thing to say on camera when your company sells AI.

All three spent the previous year hammering a very specific fear into public consciousness. Then all three backed off within the same week.

Funny how timing works.

The Numbers Say Something Different

Look, I know “historical data” sounds like a dodge when someone’s arguing about the future. But here’s the thing. We actually have three years of post-ChatGPT data now.

It’s not theoretical anymore.

U.S. civilian employment has grown 145% since 1962. Through electrification. The internet. Mobile computing.

Every single wave of automation that was supposed to end work as we know it.

Data-center construction alone added 200,000 jobs since 2022.

That’s the infrastructure feeding the AI that was supposed to kill jobs. Goldman Sachs CEO David Solomon made this point in a New York Times op-ed that Fortune picked up. He’s not selling AI software, for what it’s worth.

Yale Budget Lab got specific. They looked at actual occupational data since late 2022. The ChatGPT launch window. And found no significant changes. Not in job mix. Not in unemployment duration for exposed roles. Nothingburger.

Three years of nothingburger.

The doomsday prophets have been wrong for three years straight.

Honestly? I didn’t expect that either.

Why the Timing Should Make You Skeptical

Here’s what the coverage mostly glossed over: both Altman and Amodei are eyeballing blockbuster IPOs right now.

Altman’s S-1 landed at the SEC on May 22. His “I was pretty wrong” moment came four days later. That’s not a coincidence. That’s a corporate communications calendar.

Fortune called it out explicitly.

AI Magazine described their change of tone as a “major corporate pivot” ahead of public listings and regulatory scrutiny. When your valuation depends on regulators not crushing you and investors believing in massive future revenue, saying “don’t worry, jobs are fine” is a very specific kind of statement.

It rhymes with: please buy stock.

Deutsche Bank’s Torsten Slok made the actual economic case worth understanding. Lower cost per interaction doesn’t equal fewer interactions. If AI makes something cheaper to deliver, demand expands to fill that price reduction. That’s a real economic model. It just happens to be the optimistic one. The one that justifies a high IPO multiple.

The pessimistic narrative served them when they needed goodwill for regulatory treatment.

The optimistic one serves them now.

That’s the game. The people who profit most are usually the ones who know which story is being told to which audience at any given moment.

What Actually Moved in the Market

The revised position has internal coherence, I’ll give them that. AI automates 90% of a job. Humans do the remaining 10%. And that 10% expands to fill attention while the machine handles the rest. Amodei called it a 10x productivity multiplier. Solomon pointed to historical precedent. Electrification, digital tools. As proof that automation consistently creates net job growth.

Maybe they’re right.

The data so far supports “nothing happened yet.” Yale Budget Lab’s findings are hard to argue with.

But here’s what nobody’s talking about: fear still ran the process. Even if the underlying prediction was wrong.

Companies used the AI narrative to justify actual layoffs. Meta cut 8,000 people in May. Intuit cut 3,000. Same month. Same excuse. The people making those decisions knew the timeline kept slipping. Some of those same executives said AI was “productive and useful only six months ago” right before reversing course.

But the fear was leverageable, so it got used.

You felt that in budget meetings.

In client conversations. In the hesitation before hiring.

That fear served someone else’s interests. Yours got caught in the crossfire.

The Actual Opportunity Nobody’s Naming

Huang’s “lazy” comment stuck with me. He wasn’t wrong.

Cutting headcount is the easiest way to make quarterly numbers look better. Blaming AI lets executives off the hook for a decision they made for boring, old-fashioned reasons — they needed to hit a number. And headcount was the lever. The AI story was a cover. A convenient one.

So here’s the question nobody’s asking small operators: what do you build while the narrative catches up to reality?

If job replacement is genuinely slower than advertised. And three years of flat occupational data suggests it is. Then there’s a window. Not permanent. But longer than the fear merchants implied.

The teams moving now, building the workflows, installing the tooling, figuring out the judgment calls: those are the ones who benefit when the next capability wave lands. And there will be a next wave. That’s not the fear talking — that’s just how technology works.

The 10% that stays human is where value concentrates. Judgment. Relationship. Knowing which problem is actually worth solving. Those skills don’t automate out.

Side note: the docs for most of these AI tools are still a mess. But that’s a other problem.