Here’s what Sam Altman said before his company filed to go public: “A lot of jobs will disappear. Society is not prepared.”
Here’s what Dario Amodei said: AI will replace “the majority of knowledge work within five to seven years.” He called the labor market transition “the hardest problem we face.”
Here’s what Altman said four days after OpenAI confidentially filed an S-1 targeting a $1 trillion valuation: he was “delighted to be wrong” about AI destroying jobs. The technology, he now claimed, would create more roles than it eliminated.
Same week, Amodei told a private investor briefing that his five-year window was “overly pessimistic.” Anthropic was building AI to “augment, not replace.”
Two CEOs. Two companies with combined IPO ambitions north of $1.9 trillion. Two reversals that arrived with the precision of a Swiss train timetable.
Nobody is saying they coordinated. They didn’t need to. They just faced the same structural pressure at the same moment. You cannot file an S-1 that says “our core product replaces white-collar workers at scale” and then stand in front of institutional fund managers asking them to underwrite your growth multiple. The math doesn’t work. So the narrative gets a haircut.
The timeline tells you everything
Let’s lay out the sequence.
OpenAI’s S-1 hit the SEC confidentially in mid-May 2025 – the JOBS Act lets companies keep financials private while the SEC reviews. The public learned about it four days later through a carefully timed leak.
Altman’s Sydney speech at the International AI Summit dropped into that exact news cycle. “Delighted to be wrong” is a beautiful piece of linguistic engineering. It retroactively reframes his apocalyptic warnings as caution rather than conviction. He didn’t say “I was wrong.” He’s *delighted to be wrong* – a construction that lets him claim credit for optimism without admitting error. It’s the kind of thing a politician says, not a technologist.
Amodei’s pivot ran the same playbook. At Anthropic’s investor roadshow, he recast his five-to-seven-year knowledge work replacement forecast as “a product of extrapolating early data.” Now he saw a ten-to-fifteen-year transition with significant “human-in-the-loop requirements.” The timeline stretched by a factor of two or three – conveniently past the point where most fund managers’ holding periods expire.
This is not new
Zuckerberg spent 2010-2012 telling anyone who’d listen that Facebook’s mobile transition was a dire risk. Mobile monetization was “the biggest question mark.” Then Facebook went public. Eighteen months later it was a mobile advertising machine worth ten times its IPO price. The risk narrative was real, but it was also useful: it set low expectations.
Uber’s 2019 IPO was even closer. Dara Khosrowshahi emphasized autonomous vehicles as a distant, uncertain future – don’t worry about our core business being replaced by robots, that’s years away. Post-IPO, the messaging quietly shifted to “autonomous technology is arriving faster than expected and we’re positioned to benefit.”
Coinbase’s 2021 S-1 spent 40 pages warning that the SEC could classify most crypto as securities and shut down the whole business. The threat was real, but it also functioned as a narrative floor. When regulatory clarity didn’t arrive with the severity described, Coinbase could claim resilience.
Every disruptive company softens its existential messaging at IPO time. Not because the executives are liars. Because a public company’s fiduciary duty to maximize shareholder value conflicts directly with honest description of negative externalities. The most honest CEO in the world, when faced with that choice, shades the truth toward the share price.
Safety is a budget item, not a mission
Safety research is expensive. OpenAI and Anthropic have together spent hundreds of millions on alignment research, red-teaming, and policy advocacy. Those are pure overhead – they don’t generate revenue, expand markets, or improve unit economics.
OpenAI reported roughly $25 billion in revenue in 2025 against about $14 billion in losses. Negative 56% margin. Anthropic hasn’t disclosed similar numbers, but analysts estimate comparable ratios on a smaller scale. Both are burning capital at a pace that makes public market access not optional but existential.
Pre-IPO, safety-first messaging attracts top research talent and favorable regulatory treatment. It helps you argue for compute access, data rights, and policy carve-outs. Post-IPO, safety spending becomes a line item that depresses margins. The narrative must shift from “we are the responsible ones” to “we are building at scale.”
The same executives who testified before Congress about existential risk are now telling investors the risk is manageable, distant, and priced in. Neither statement is obviously false. But the direction of the shift tracks the direction of the fundraising cycle with a correlation that should trouble anyone who uses these forecasts to make decisions.
What to do with this information
The takeaway is not that Altman and Amodei are dishonest. It’s that CEO narratives about the future must be read with an explicit discount for the speaker’s current capital-markets position.
A pre-revenue founder who tells you AI will eliminate every job is telling you the truth as he sees it. That same founder, four days after filing an S-1, is telling you something different – not because the technology changed, but because his audience changed.
Before taking any CEO prediction at face value, ask three questions:
**What is this company’s capital-markets calendar?** If they’re in a quiet period, roadshowing an IPO, or negotiating a secondary offering, every statement is a marketing document.
**What would this person have to believe to make their current statement consistent with their previous one?** Altman needs you to accept he was “delighted to be wrong.” Amodei needs you to believe the timeline tripled in six months without any major technical breakthrough.
**Who benefits from the shift?** The underwriting syndicate, the pre-IPO investors seeking liquidity, and the employees holding vested options. The retail buyer entering at the IPO price is the last person the narrative is designed to serve.
The $2 trillion question is not whether AI will destroy jobs. It’s whether the public markets will price in the truth before the lockup periods expire. Historical precedent suggests they will not.
The pivot is designed to work just long enough.