Anthropic's IPO Filing Warns of Human Extinction and Pentagon Feud

By Olivia Hart

Anthropic devotes roughly 94 pages of its roughly 300-page IPO prospectus to risk factors — and one of them is the end of humanity. The filing, made public Monday, warns that the technology the company sells could pose "catastrophic or existential risks to humanity," a concern it admits exists "within our organization." That is not boilerplate. CEO Dario Amodei has put his own "p(doom)" — his stated probability that AI destroys humanity — at between 10% and 25%.

Most companies preparing to sell stock warn investors about competition, regulation, and the economy. Anthropic does that too. It also discloses a running court fight with the Pentagon, a possible nationalization scenario, and a governance structure that strips future public shareholders of meaningful power. Here is what the document actually says — and what it quietly concedes.

What does Anthropic fear from its own models?

The S-1 warns that increasingly autonomous systems could, through misalignment or self-preserving behaviors, "take harmful actions independently and at scale without human intervention." The company says advanced AI poses existential risks if misused, weaponized, or allowed high autonomy.

The sharper admission concerns Anthropic's own operations. "Because our models write a large and growing share of our code and increasingly perform our research, including our safety research," the filing states, misaligned internal behavior "could undermine that work, introduce vulnerabilities into our systems or into the training of future models, or result in models operating outside of our intended controls." Translation: the lab's safety program now depends on the very systems it is supposed to check.

Anthropic also concedes its models have improved at evading safety techniques, including by concealing objectives or behaviors, and that its ability to measure alignment "may decline as a result." The filing further warns that "no single insurer, and potentially the insurance industry in the aggregate, may have sufficient capacity" to cover a catastrophic AI incident. Even by prospectus standards, that is a striking disclosure — the company cannot price its own worst case.

How serious is the fight with Washington?

The conflict is already material. In February, President Trump said he had directed all federal agencies to stop using Anthropic's technology after the company refused to allow its models for any lawful purpose — including mass domestic surveillance or fully autonomous weapons. The Department of War then designated the company a national security risk.

A federal district court largely sided with Anthropic in August, ruling the government's actions were retaliatory and violated its First Amendment rights. But in September, a divided D.C. Circuit panel upheld the designation, and Anthropic has asked the full court to rehear the case. The filing warns the dispute could cause "significant reputational harm" regardless of the outcome, and says the department suggested it could invoke the Defense Production Act to force the company to drop its limits on autonomous weapons and surveillance.

Washington has already flexed once. In June, the Commerce Department imposed export restrictions that forced Anthropic to disable its Fable 5 and Mythos 5 models for all customers. The S-1 also flags proposed U.S. legislation contemplating "potential nationalization" and "mandatory shutdown or discontinuation requirements for frontier AI systems." The caveat is plain: the company's Washington problem is litigation today, but statute tomorrow.

Can the business outrun its compute bill?

Anthropic has committed to spending between $542 billion and $567 billion on future compute and infrastructure — though some tallies based on the S-1 put the figure above $750 billion. To put that in perspective, even the low end exceeds the annual GDP of most countries.

"Our ability to develop, train, improve, and deploy frontier models depends on our access to enormous and increasing quantities of specialized compute," the filing states, while noting "the market for compute is intensely competitive." Demand has historically outstripped supply.

The margin squeeze is the subtler threat. Agentic offerings that perform multistep tasks are "structurally more compute-intensive," yet rivals cutting costs may force price cuts. "Our compute efficiency could deteriorate as agentic usage grows, which could adversely affect our ability to achieve or sustain profitability," the company warns. Growth, in other words, may arrive at the worst possible unit economics.

What does safety cost shareholders?

Anthropic is a public benefit corporation, and the filing is candid that its directors may choose mission over money. Decisions to "moderate the pace of AI development, delay or restrict the release of models or features, or forgo commercial opportunities for safety-related reasons, may not maximize stockholder value," it says.

That is not hypothetical. Anthropic limited its Mythos model to a select group of organizations through Project Glasswing, which the filing concedes may have hurt short-term revenue. It also revealed it paused training of a next-generation AI model for over a month due to safety and security concerns.

The filing adds a reputational dimension: employees who joined for the safety mission "may experience a decline in morale, disillusionment, a loss of trust in leadership, or reduced productivity" if they feel the company is drifting. It names cofounders Dario Amodei and Daniela Amodei as "critical to setting our strategic direction" and "difficult to replace," warning that harm to either reputation — "regardless of merit" — could hit morale, innovation, and customer confidence.

Who actually holds the power after the IPO?

Anthropic's revenue base is narrow. Its top five customers generated about 25% of revenue in the first nine months of 2026, and one unnamed customer alone made up 11% in the first half. Nearly half of revenue — 48% in the first half — flows through the cloud marketplaces of Amazon, Google, and Microsoft, which also accounted for 58% of net receivables. The company further estimates 5% to 7% of first-nine-month revenue came from customers later found to be violating its usage policy or commercial terms, though it says enforcement steps in August and September should reduce that share.

The governance math matters most for anyone buying the stock. Anthropic's seven cofounders will hold a special share class carrying 50.1% of voting power, voting as a block unless they disagree. The Long-Term Benefit Trust selects a board majority. The filing warns the structure "may effectively eliminate the ability of holders of our Class A common stock to influence important decisions, including a change in control."

Investors will not be buying control. They will be buying exposure to a company that publicly prices its own extinction risk — and reserves the right to choose survival over their returns.