AI Doomsday Warnings Won't Stop IPOs: The Surprising Risk

AI doomsday warnings have dominated September, and they arrived at the worst possible moment for the two companies at the centre of them. Anthropic is expected to list later this year at a valuation that could approach $2 trillion, and OpenAI plans to follow in 2027. In the same fortnight, a researcher resigned from Anthropic warning that the industry was “gambling with our lives”, Anthropic’s own chief executive called for AI development to slow down, and OpenAI’s chief executive called this an “ill-advised moment” to go public.

The consensus on Wall Street, as an AFP report put it on 18 September, is that the warnings are unlikely to slow the IPOs. The questions that linger are more specific: how a company that says its product might be catastrophically dangerous describes that risk to investors, how the market prices it, and whether regulators should step in first.

This article sets out the timeline, the arguments on each side, and the disclosure rules that will force the issue into writing. The prospectus, not the podcast, is where the AI doomsday debate meets securities law.

Table of contents

The Fortnight of AI Doomsday Warnings
Why AI Doomsday Warnings Collide With an IPO
What the SEC Actually Requires in Risk Factors
The Bull Case: Markets Can Price AI Doomsday Risk
The Bear Case: AI Doomsday Talk as Marketing
What the Companies Say About AI Doomsday Risk and Growth
Valuations Under the AI Doomsday Debate
Five Questions an AI Doomsday Prospectus Must Answer
Should Regulators Act on AI Doomsday Risk Before the IPOs?
What Investors Should Watch in the AI Doomsday Debate
AI Doomsday Warnings and IPOs: The Verdict
References

The Fortnight of AI Doomsday Warnings

The AI doomsday warnings came in quick succession, and several of them came from inside the companies preparing to sell shares.

The resignation

Jacob Coxon, a researcher who had previously worked at OpenAI, left Anthropic in the second week of September. “The people building AI earnestly believe that it could kill us all by the end of the decade,” he wrote. AFP reports that former colleagues rallied round him, including one who voiced concern about a small probability that AI could cause humanity’s extinction. We examined how rarely such language appears in Anthropic’s own published safety work in our extinction risk analysis.

The essay

Days later, on Saturday 12 September, Anthropic chief executive Dario Amodei published an essay arguing that frontier development must be paced. He wrote of his “worry that in 6-12 months” a swarm of AI agents “could be capable of taking over the entire internet”, potentially causing “hundreds of billions of dollars in damage”. AFP notes that the essay “didn’t say anything about the IPO”. Our coverage of pacing the frontier covers its proposals.

The chorus

Elon Musk agreed with Amodei, and OpenAI’s Sam Altman backed the call too, warning that without safeguards humans could “lose control of the future of AI”. Google DeepMind’s Demis Hassabis said the direction was right. On the same weekend Altman said OpenAI would not go public in 2026, calling it an “ill-advised moment”.

Date (2026)	Event	Source
8 to 9 Sep	Jacob Coxon resigns from Anthropic with an extinction warning	Intelligencer, AFP
12 Sep	Amodei essay calls for pacing frontier AI	IBTimes, AFP
12 Sep	Altman: OpenAI IPO “not 2026”, an “ill-advised moment”	Fortune via AFP
14 Sep	Michael Burry calls the slowdown push “self-serving”; Trump attacks the “conspiracy”	IBTimes, Intelligencer
15 Sep	OpenAI CFO Sarah Friar: “Even if we stop today…”	CNBC via AFP
16 Sep	Anthropic’s Sarah Heck: safety “the core of who we are”	Politico conference via AFP
17 Sep	Steve Eisman: “Postpone your IPO”	CNBC via BigGo Finance
18 Sep	AFP: AI doomsday warnings unlikely to slow IPOs	AFP via The Economic Times

Why AI Doomsday Warnings Collide With an IPO

A private company can say almost anything about the future of its technology. A company selling shares to the public cannot, and that is the collision the AFP report identifies.

The disclosure duty

Companies must disclose known material risks before an IPO. AFP notes that Anthropic could submit its public registration document to the Securities and Exchange Commission as early as this month, and that “it’s unclear if, or how, either company will address theoretical doomsday scenarios in securities filings.”

The confidential start

Anthropic submitted a confidential draft registration statement on 1 June 2026, and OpenAI announced its own confidential filing on 8 June, as our coverage of the Anthropic IPO timeline and the OpenAI IPO delay set out. A confidential draft lets a company work through SEC comments in private. It has to be made public before the marketing roadshow begins.

Why this time is different

Plenty of companies list risks that sound dramatic: pandemics, cyberattacks, climate events. What is unusual here is that the company’s own leadership is the loudest source of the AI doomsday warning. A risk that the chief executive describes in an essay is hard to call remote in a prospectus.

What the SEC Actually Requires in Risk Factors

The rule that governs this is Item 105 of Regulation S-K, and its wording matters more than any podcast.

The core test

Item 105 requires a discussion of “the material factors that make an investment in the registrant or offering speculative or risky”. Materiality, not probability alone, is the test: a low-probability risk with an enormous consequence can still be material.

The 2020 changes

Amendments in 2020 added two relevant requirements. If the risk-factor section runs longer than 15 pages, the company must include a summary of no more than two pages at the front of the document. And companies are discouraged from generic risks that could apply to any issuer, which pushes them to describe the risks specific to their own business.

How that applies to an AI lab

A generic line such as “AI may be subject to regulation” will not capture what Anthropic’s chief executive has said publicly. Specific risks could include a regulatory pause on frontier training, liability for misuse, a voluntary slowdown the company itself advocates, and the reputational effect of staff resignations. Each can be written as a business risk without taking a view on extinction.

Public statement	How it could translate into a risk factor	Business effect investors would price
Amodei: development must be paced	We may voluntarily slow model releases	Slower product cadence and revenue growth
Calls for binding safety rules	Governments may restrict frontier training or deployment	Compliance cost; possible pause
Agent swarm “taking over the entire internet”	Our models could be misused for cyberattacks	Liability, insurance and reputational cost
Staff resignations with public warnings	We depend on retaining key researchers	Talent loss and hiring cost
Nonprofit or public-benefit control	Our governance may prioritise safety over shareholder returns	Limited shareholder influence

What we do not know

Neither company has published a registration statement, so nobody outside knows how they have drafted these sections. The table is an illustration of how public statements map onto the categories Item 105 asks for, not a prediction of the text.

The Bull Case: Markets Can Price AI Doomsday Risk

The most prominent investors quoted in the coverage think the market can absorb the AI doomsday warnings, and some think a listing will help.

Brad Gerstner

Altimeter Capital founder Brad Gerstner, who holds shares in both OpenAI and Anthropic, argued that an IPO brings transparency. “Anthropic will IPO. The market knows how to price risk, see SpaceX,” he posted on X. “There is huge appetite to invest in the AI leaders.”

Accountability as a feature

Lise Buyer of Class V Group told CNBC that “sooner is better than later for a public offering as the accountability that comes with being a public company might be of a great interest to many.” Gartner analyst Arun Chandrasekaran suggested stricter standards “could actually favor Anthropic and OpenAI” if smaller competitors cannot afford the required safety investment.

Prediction markets

Traders are not deterred by the AI doomsday talk either. BigGo Finance reports that Polymarket gave an 87 per cent probability to an Anthropic IPO by 31 December, with about $3.2 million traded on that contract, and 73 per cent to Anthropic closing its first day at a market capitalisation of at least $2 trillion.

Polymarket probabilities on Anthropic’s listing, as reported 17 Sep 2026

IPO by 31 December 87%

First-day market cap at least $2tn 73%

Market cap $2tn to $2.25tn 27%

Market cap $2.25tn to $2.5tn 25%

Prediction-market prices move by the minute and reflect traders’ bets, not forecasts from the companies. They are shown as reported on the day.

The Bear Case: AI Doomsday Talk as Marketing

The sceptics’ argument is not that AI is safe. It is that AI doomsday warnings serve the companies issuing them.

Chamath Palihapitiya

The All-In podcast co-host questioned the logic on air: “Are we to believe that there is something that’s extremely dangerous that’s hiding inside this company because this person that quit said it, and then it was amplified by a bunch of people” who still work there? And if it is true, he said, investors “will demand an enormous discount.”

Michael Burry

In a post on X on 14 September, Burry called the push to slow frontier AI “self-serving”, IBTimes reports. His four points: the large language model is not AI, so “there is nothing AI to slow down”; rivals are “coming up fast” so a slowdown helps incumbents; IPOs need “hype & puffery”; and safety talk offers “cover for real uncontrollable slowing growth as IPOs look to be pushed out.”

Steve Eisman

“I think this whole Terminator thing is garbage. Just garbage,” Eisman told CNBC on 17 September, according to BigGo Finance. “Put your money where your mouth is. Postpone your IPO.” He argued that “token maxing is over, the open-weight models are taking big market share,” and that the labs “realize that there are no moats around their business whatsoever.”

Voice	Position on the warnings	Implication for the IPOs
Brad Gerstner, Altimeter	Real risks, priceable	“Anthropic will IPO”
Lise Buyer, Class V Group	Listing adds accountability	Sooner is better
Chamath Palihapitiya	Sceptical of the resignation story	If true, an “enormous discount”
Michael Burry	“Self-serving” hype	Cover for slowing growth
Steve Eisman	“Garbage”	“Postpone your IPO”
Donald Trump	A “SICK conspiracy” against AI	“Don’t kill the Golden Goose!”

The market’s own verdict that day

IBTimes notes that AI-linked stocks fell on Monday 14 September, with the Nasdaq 100 down more than 1 per cent in early trading. If the AI doomsday narrative were designed to inflate the sector, the listed market was not cooperating.

What the Companies Say About AI Doomsday Risk and Growth

Both companies have tried to separate AI doomsday safety concerns from the business case, in different ways.

OpenAI: the business survives a pause

OpenAI’s chief financial officer Sarah Friar dismissed the idea that slowing the release of state-of-the-art models would hurt revenue. “Even if we stop today, the amount of intelligence that’s available in the world is massive,” she told CNBC. Earlier, at an all-hands meeting, she told staff OpenAI “will be a public company in 2027.”

Anthropic: safety is the brand

Anthropic’s head of policy Sarah Heck rejected the idea that being safety-minded would compromise the IPO. “I would say safety has been the core of who we are from the very beginning,” she said at a Politico conference. “Our investors know that. Our customers know that.”

The tension neither resolves

Friar’s line implies the business does not depend on further capability gains; the valuations imply that it does. Heck’s line implies safety is priced in; Amodei’s essay implies the risk is rising. Investors will want to see which version the risk factors describe.

OpenAI’s dealmaking continues

AFP reports that OpenAI is in talks to raise new funding at a valuation of at least $1.2 trillion before going public, and that leaders were concerned about investor skittishness, market uncertainty and SpaceX’s share performance. The AI doomsday debate has not stopped the private fundraising.

Valuations Under the AI Doomsday Debate

The numbers explain why nobody is postponing over AI doomsday fears. Each figure below is as reported; none is a filed price range.

Reported valuations, US$ trillions

Anthropic, possible IPO valuation ~2.0

OpenAI, new round under discussion 1.2+

Anthropic, May 2026 round 0.965

OpenAI, March 2026 round 0.852

Bar widths are each valuation divided by $2 trillion.

The step-ups

A $2 trillion listing would be 2.07 times Anthropic’s $965 billion May round, which BigGo describes as “nearly double”. OpenAI at $1.2 trillion would be 1.41 times its $852 billion March round. Reuters has reported, via BigGo, that Anthropic could raise as much as $100 billion.

The SpaceX comparison

Gerstner’s “see SpaceX” cuts both ways. AFP says SpaceX raised a record $85 billion in its June IPO, but its stock has lost about a quarter of its value since peaking at around $202 a share. A quarter off $202 is roughly $150. The market priced the risk, and then kept pricing it.

Who else is exposed

AFP notes that Microsoft, Amazon, Google and Nvidia all hold significant stakes in the two labs, and that the US economy is increasingly tied to the AI buildout. How the Anthropic IPO performs will also set the tone for OpenAI’s debut next year. See our Nasdaq listing coverage for the exchange choice.

Five Questions an AI Doomsday Prospectus Must Answer

Whatever the final wording, the registration statement will have to take a position on a handful of questions that the essays and podcasts have left open. These are the ones investors and analysts are likely to press.

Is the AI doomsday risk material, or remote?

The company’s own leaders have described catastrophic outcomes as plausible within years. A prospectus that treats AI doomsday scenarios as too remote to mention invites the question of why the chief executive wrote an essay about them. One that treats them as material has to explain how the business survives them.

Would the company slow down if it had to?

Amodei has argued for pacing the frontier. Investors will want to know whether that is a commitment, a preference or a sentiment, and what triggers it. A voluntary slowdown is a business decision with revenue consequences, and it belongs in the risk factors if management would actually take it.

Who decides?

Both labs have unusual governance: OpenAI’s nonprofit foundation appoints its group board, and Anthropic operates as a public benefit corporation with safety-focused oversight. The prospectus must explain who can overrule commercial management on safety grounds, because that is the practical answer to how AI doomsday concerns would be acted on.

What happens to revenue in a pause?

Friar’s claim that the existing models are valuable “even if we stop today” is testable. A prospectus that includes segment revenue, customer concentration and contract terms will let analysts estimate how much of the business depends on the next model rather than the current one.

How will the company talk about this after listing?

Public companies face ongoing disclosure duties, and executives’ public statements are scrutinised against filings. After listing, every new AI doomsday warning from a senior employee becomes a potential market-moving event. Investors will want to know whether the company has a policy for that, and so will its lawyers.

Should Regulators Act on AI Doomsday Risk Before the IPOs?

AFP frames the third lingering question as whether regulators should step in. There are two quite different kinds of regulator involved.

The securities regulator

The SEC does not judge whether a business is wise, only whether it is disclosed. Its leverage is the comment process on the registration statement, where staff can ask a company to expand or sharpen a risk factor. If Amodei’s public statements are material, the SEC can ask why the prospectus does not reflect them.

The AI regulators

Binding AI safety rules are a separate track, and the political signals point in opposite directions. Amodei, Altman and Musk have called for coordinated safeguards. President Trump posted that “the only control or ‘guardrail’ that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT,” and warned: “Don’t kill the Golden Goose!”

Why the disclosure route matters

Even without new AI law, a prospectus creates legal accountability. Statements in a registration statement carry liability if they are materially misleading. That is a sharper constraint on how a company talks about AI doomsday risk than any essay, and it is why the drafting of those pages will be read closely. Amodei’s worry about agent swarms is, in business terms, a cybersecurity and liability risk, and that is language a prospectus can hold.

What Investors Should Watch in the AI Doomsday Debate

The debate will resolve into documents and dates over the next few months.

The public filing

When Anthropic’s registration statement becomes public, read three places: the risk-factor summary, the governance section on who controls safety decisions, and any discussion of voluntary slowdowns. Compare them with the language in Amodei’s essay.

The roadshow

Management will face investor questions on the AI doomsday debate directly, probably in every meeting. Watch whether the answers match the filing, and whether the company commits to any external safety commitments that would bind it after listing.

For businesses buying AI

The same AI doomsday risks affect customers, indirectly. A voluntary pause or a regulatory restriction could change model availability, pricing or roadmaps. Build contracts with notice periods and portability, and avoid dependence on a single frontier model. Our AI strategy services cover how to plan for that, and the AI models, tools and releases hub tracks the providers.

AI Doomsday Warnings and IPOs: The Verdict

The AFP headline is probably right: AI doomsday warnings are unlikely to slow the IPOs. Prediction markets, prominent backers and the companies themselves are all behaving as though the listings will happen, and the valuations give every reason to proceed.

The surprising risk is not that AI doomsday talk derails a listing. It is that it becomes a disclosure problem. A company whose chief executive says in public that the technology could cause enormous damage within a year has to decide how to say that to investors in a document that carries legal liability. Too little, and the prospectus looks misleading. Too much, and, as Palihapitiya put it, investors “will demand an enormous discount.”

Between those two failures lies the drafting task of the year. The sceptics will read the risk factors as a test of sincerity; the bulls will read them as a test of pricing. Both will be reading the same pages.