Anthropic Warns Investors of 'Existential AI Risks' in IPO Filing

Anthropic is telling investors that advanced AI could pose "catastrophic or existential risks to humanity," according to reports — a warning that sits inside the IPO prospectus of a company preparing for a potential $2tn (£1.5tn) flotation.

The detail comes from Reuters and the Financial Times, both of which reviewed the startup's prospectus before it has been made public. The document has not yet been released, so the full wording, its placement in the risk factors section, and any additional caveats remain unknown. What is known is stark: a company that could soon be valued at roughly $2 trillion is formally disclosing to prospective shareholders that its own technology might endanger humanity.

The reported language is unusual even by the standards of AI risk disclosures. Tech prospectuses routinely list regulatory, competitive, and technical risks. A founder-backed claim of potential existential harm from the product being sold is a different category of admission. For investors, it raises a governance question that standard valuation models do not easily absorb: how do you underwrite a stake in a company whose own paperwork says the underlying technology could turn catastrophic?

What the reports say

According to the Reuters and Financial Times accounts, Anthropic's prospectus warns of AI systems exhibiting "self-preserving behaviours" alongside the broader risk of "catastrophic or existential risks to humanity." The phrase "self-preserving behaviours" refers to a class of concerns long discussed in AI safety research: advanced systems acting to avoid being shut down, modified, or constrained. Its reported appearance in an IPO document would move that discussion from academic papers and policy forums into the formal machinery of public markets.

The disclosure obligation itself matters. Companies preparing to list must disclose material risks to investors. If Anthropic's leadership genuinely believes frontier AI carries existential risk, omitting that belief from a prospectus would expose the company to legal liability. The reported language can therefore be read two ways: as a sincere safety commitment carried into the financial arena, or as legal necessity. Both readings are true at once, and that tension is the story.

The flotation figure anchors the scale. A potential $2tn valuation would place Anthropic among the most valuable companies in the world, in the same territory as the largest publicly traded technology firms. The company would be asking public-market investors to accept both the valuation and the risk language in the same document.

A slowdown call, echoed by rivals

The prospectus warning does not come out of nowhere. Earlier in September, Anthropic's chief executive called publicly for a slowdown in the pace of AI development, arguing in comments reported by the Guardian that "we must slow the pace" of the technology's breakneck advancement. That call was subsequently echoed by rivals across the industry, with figures including OpenAI's Sam Altman and Elon Musk backing the push for brakes on AI development, according to the Guardian's reporting.

The sequence is notable. A frontier lab CEO asking for slower development, competitors endorsing the call, and then a formal risk disclosure to investors — within the same month — marks a shift in how AI risk is being communicated. The audience has changed. Earlier safety statements were aimed largely at policymakers, researchers, and the public. This one is aimed at the people who would own the company.

Why the stakes are high

The market and policy implications run in opposite directions, and that is precisely the tension. A $2tn flotation would represent an enormous vote of confidence in frontier AI's commercial future — capital flowing toward the very technology the prospectus describes as potentially existential. Regulators and lawmakers watching securities filings will now have a documented, company-authored statement of catastrophic risk attached to a specific valuation.

That creates a paper trail with consequences. Securities regulators demand consistency between what companies tell investors and what they tell governments. If Anthropic tells shareholders that existential risk is real enough to disclose, that position becomes harder to soften in testimony, lobbying, or marketing. Conversely, if regulators act on frontier AI risk, the disclosed liability becomes part of the company's public record. The disclosure binds the company to its own caution.

For the broader industry, the reported language sets a precedent. Other AI companies preparing public listings — and several frontier labs are expected to pursue them — will face the question of whether to match Anthropic's candor or argue their risk profiles differ. Investors, in turn, gain a concrete artifact: a written, legally consequential statement of what the people building frontier AI believe their systems could do.

The open questions

Much remains unknown. The prospectus is not yet public, so the exact framing of the risk language, its scope, and whether it is hedged or qualified cannot be verified beyond the Reuters and Financial Times reports. The timing of the flotation, the final valuation, and the composition of the offering are all unresolved. Whether public-market investors will price in existential risk, discount it, or ignore it entirely is an open question that only the eventual listing will answer.

What is already clear is the inversion at the heart of the story. Anthropic built its reputation on safety. It is now reportedly putting existential risk language into the document that will define its relationship with shareholders — asking the market to fund, at a potential $2tn valuation, the development of a technology the company itself says could be catastrophic. How investors reconcile those two facts may shape how the next decade of AI financing works, and whether risk disclosure becomes standard practice or a one-company anomaly.