The labour-market commentariat does not often get to watch two of its loudest doomsayers retract their forecasts on the same week. On May 26 in Sydney, OpenAI CEO Sam Altman told a Commonwealth Bank of Australia conference that he had been “pretty wrong” about AI’s social and economic impact and was “delighted to be wrong.” Three days later, Anthropic CEO Dario Amodei — who had spent 2024 and 2025 warning that AI could erase half of entry-level white-collar jobs in five years — pivoted to a more genteel formulation in which automation “expands the work people do.”
These are the same week. So are the IPOs.
What changed in the narrative, and what didn’t
Altman’s prior position, repeated in earnings calls and conference keynotes through 2025, was that AI would “probably replace most of the jobs people do today” and that entire categories would be “totally, totally gone.” Amodei’s January 2025 Axios interview put a number on it: 50% of entry-level white-collar jobs erased within five years. The two CEOs ran the most-quoted half of the AI doom canon for two years and got cited in nearly every layoff press release through May 2026.
What both of them said this week is, in plain language, the opposite. Altman now believes he over-estimated the economic dislocation. Amodei now believes automation expands work. Goldman Sachs CEO David Solomon joined the chorus the same week.
What did not change in the same week: Cloudflare cut 1,100 jobs and disclosed internal AI usage was up 600% in three months. Coinbase cut 14% and said AI was the reason. BILL announced cuts of up to 30%. Upwork announced cuts of roughly a quarter. Amdocs announced 3,000 cuts — its first round formally labelled “AI-era restructuring.” May 2026 looks set to close around 30,000 net tech jobs lost, with roughly half explicitly attributed to AI on the public record.
So: the operators kept laying people off. The vendors of the technology decided this was a bad time to be quoted in the press releases.
The IPO clock
OpenAI’s confidential S-1 filing was reportedly submitted in May 2026, at a target valuation in the trillion-dollar zone. Anthropic completed a secondary sale at a $300B mark in March 2026 and is widely reported to be teeing up its own IPO at a comparable valuation. Both founders are in the live preparation window where every public statement is a draft of prospectus language and every line of conference video is a candidate for a future short-seller report.
A roadshow tested with retail and pension-fund money does not want to begin with the question “is your product responsible for the headlines about layoffs?” — particularly when the answer in 2025 was an unembarrassed yes. Roadshow sentiment is also an empirically priced variable: Reuters surveyed retail investors in May 2026 and found a 9% discount priced into companies “associated with mass displacement narratives.” 9% of a trillion-dollar valuation is $90B. The cheapest way to recover $90B in valuation is to revise the talking points.
Both founders did. The same week.
What the regulators were doing while the narrative reset
California Governor Gavin Newsom signed an executive order on May 21 directing the California Labor and Workforce Development Agency to evaluate severance standards, expand UI enrolment, and reform the state’s WARN Act specifically for AI-driven displacement, with a 180-day deadline for recommendations. The EU AI Act’s employment-impact disclosure requirements are now in force. New York’s amended WARN Act started requiring AI-related disclosure on layoff notices in 2025; Hunton’s tracking notes that companies have so far declined to file AI as the cause on a single notice in NY.
The regulatory direction of travel is firmly toward attaching specific liability — severance, unemployment insurance, notice — to AI-attributable cuts. A foundation-model CEO whose own video clips are quoted in newsroom layoff coverage becomes, as a matter of literal evidence, easier to subpoena. The fastest way to make that evidence less quotable is to stop saying the thing.
What this changes for workers, in practice
Very little, on the cuts themselves. The decision to lay off the customer-support team and replace it with Salesforce’s Agentforce or Cisco’s automation suite was not made because Altman said the displacement was coming. It was made because, at the operator level, the unit economics finally penciled out. The narrative that justified the press releases was useful when it was bullish for the buyers’ procurement budget and bullish for the sellers’ valuation. As soon as the same narrative becomes a liability for the sellers’ IPO, it gets replaced. The cuts don’t.
The likely operating-level shifts, on the other hand:
- Vendor press packets will quietly drop “AI replaces work” and reach for “AI augments productivity” — the same product, the same demo, recoded for the prospectus. Watch new OpenAI and Anthropic case studies for the change of register.
- The first wave of AI-as-the-reason layoff press releases — Cisco, Cloudflare, Coinbase, Amdocs, Wix — was effectively endorsed by founder-class statements. The next wave will be marketed without that endorsement. The operators may stop saying it too, particularly if California’s WARN reforms attach specific severance liability.
- Workforce-planning consultancies that have been selling AI-displacement training programs on the strength of Amodei’s 50% quote now have a new sales objection. Their next deck has to defend itself without him.
What to watch
- Whether OpenAI’s actual S-1 filing — when it becomes public — uses the word “displacement” anywhere in its risk-factors section, or studiously avoids it. Either is informative.
- Whether the layoff-tracker numbers in June and July reflect any deceleration. If they don’t, the gap between founder rhetoric and operator behavior becomes a story of its own.
- Whether Newsom’s California WARN reform, due 180 days from May 21, ends up attaching named-AI-vendor liability to displacement cuts. That would put OpenAI and Anthropic in a position where their prior statements are admissible evidence at trial. The walk-back, in that case, looks less like a vibe shift and more like a defense brief.
- Whether competitor founders — Mistral’s Arthur Mensch, xAI’s Elon Musk, Google’s Demis Hassabis — pick up the new tune or stick with the old one. The first one who keeps saying 50% out loud becomes the convenient quote for everyone else.
Altman and Amodei spent two years telling the world the jobs apocalypse was coming. The cuts came. Then, in the week their accountants drafted prospectuses, the two of them decided the apocalypse was a misunderstanding. Neither thing rolls back the other. The workers who lost their jobs in May did not get their jobs back when Altman said he was “pretty wrong.” The IPO bankers got a cleaner deck. That is the only thing that changed.