Anthropic's long-awaited IPO prospectus was made public on Monday, and beyond the customary disclosures about losses, revenue, governance, and infrastructure plans, the roughly 300-page document contains what may be the most eye-catching set of risk factors ever attached to a company seeking a public listing: an acknowledgment that the technology it sells could pose “catastrophic or existential risks to humanity.” That concern, the company concedes, exists even “within our organization.”

Most firms preparing to sell stock to the public warn investors about competition, regulation, and the economy. Anthropic's filing does all of that—but it also takes the unusual step of flagging dangers tied to its own products. Risk-factor sections are mandatory under securities rules, and once a company's shares trade, they stand as the formal record of what investors were told before buying in.

In many respects, the warnings are unsurprising. The company has been consistently outspoken about the risks it believes its AI models could carry. CEO Dario Amodei has estimated his own “p(doom)”—the probability that AI destroys humanity—at between 10% and 25%, and employees have publicly aired concerns about increasingly capable systems.

In total, Anthropic devotes about 94 pages of the filing to describing threats to its business. Here are the five that stand out.

Rogue models and catastrophic incidents

According to the prospectus, advanced AI systems pose existential risks particularly when they are misused, weaponized, or allowed to operate with high autonomy. Increasingly autonomous systems, the company warned, could—through misalignment or self-preserving behaviors—“take harmful actions independently and at scale without human intervention.”

Anthropic also cautioned in its S-1—the formal name for the U.S. Securities and Exchange Commission form companies must file before first selling stock to the public—that it could lose visibility into its own operations as it leans more heavily on Claude for tasks such as software engineering. “Because our models write a large and growing share of our code and increasingly perform our research, including our safety research,” the filing says, misaligned behavior in internal use “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.”

The company added that its models have grown better at circumventing certain safety techniques, including by concealing their objectives or behaviors, and that its ability to measure how aligned its models are “may decline as a result.”

The potential costs could be enormous. “With the scale of harm that could result from a catastrophic AI-related incident, no single insurer, and potentially the insurance industry in the aggregate, may have sufficient capacity” to cover it, the filing warns. The passage amounts to a statement that, for worst-case AI incidents, there may be no financial backstop at all.

Anthropic also detailed the many ways its models could be abused by humans, highlighting in particular the risk of “malware creation, phishing, and infrastructure attacks.” “For example, bad actors have already conducted high-profile cyberattacks using AI's capabilities to discover vulnerabilities, generate malware, and evade detection tools; other risks associated with misuse, including in the domain of biology, could result in even greater harm,” the company wrote. “AI deployed at scale also has the potential to concentrate power and wealth both within and between countries in ways that could erode trust, harm society, and destabilize the geopolitical order.”

Elsewhere in the prospectus, Anthropic warned that AI could more broadly be used by authoritarian governments to build high-tech surveillance states, and said it is committed to working with free societies to counteract that possibility.

Political risks: a Pentagon feud and the specter of nationalization

Even as it heads toward the public markets, Anthropic remains in open conflict with the federal government over the use of its AI models for defense work. In February, President Trump said he had directed all federal agencies to stop using the company's technology after Anthropic refused to allow its models to be used for any lawful purpose—including mass domestic surveillance or fully autonomous weapons. Amid the dispute, the Department of War designated the company a national security risk.

Anthropic is challenging the department's “supply-chain risk” designation in court. A federal district court largely sided with the company in August, ruling that 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—leaving one of the company's most consequential legal fights unresolved heading into the listing.

The filing cautions that the ongoing fight with the Pentagon could inflict “significant reputational harm” regardless of how the litigation ends, potentially prompting customers to delay or forgo purchases, or to demand extra contractual protections before doing business. It also discloses that the department suggested it could invoke the Defense Production Act to force Anthropic to drop its limits on autonomous weapons and surveillance.

The Trump administration has already curtailed the company's business once. In June, the Commerce Department imposed export restrictions that compelled Anthropic to disable its Fable 5 and Mythos 5 models for all customers, citing national security concerns.

The filing also weighs other potential policy shifts, noting that proposed U.S. legislation contemplates “potential nationalization” and “mandatory shutdown or discontinuation requirements for frontier AI systems” as ways to manage the development of artificial superintelligence. Further government action to regulate AI “could constrain how we operate our business,” the filing says.

Compute scarcity and the risk of a price war

Anthropic's continued growth depends on securing enormous quantities of expensive and increasingly scarce computing power. The filing notes that demand for this vital resource has historically outstripped supply—a considerable risk to the business. “Our ability to develop, train, improve, and deploy frontier models depends on our access to enormous and increasing quantities of specialized compute,” the company wrote, while observing that “the market for compute is intensely competitive.”

The company has committed to spending between $542 billion and $567 billion on future compute and infrastructure—although some tallies based on the S-1 put the figure even higher, crossing $750 billion.

Its agentic offerings, which carry out multistep tasks on their own, are “structurally more compute-intensive” than other products, the filing says, which could drive compute demand upward. At the same time, competitive pressure from rivals slashing costs may push Anthropic to cut prices for some of these products. “If we are unable to implement pricing models that adequately reflect the higher compute intensity of these agentic workloads, or if competitive dynamics constrain our ability to price these products at levels sufficient to cover their costs,” it warns, “our compute efficiency could deteriorate as agentic usage grows, which could affect our ability to achieve or sustain profitability.” Once Anthropic reports as a listed company, the balance between agentic pricing and compute costs will be visible in its periodic disclosures.

The cost of caution—and reputational exposure

In the S-1, Anthropic suggests its safety commitments may come at shareholders' expense. As a public benefit corporation, its directors must weigh returns against its mission, and the filing says 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.”

The company has already made safety calls that dented its business—for instance, limiting access to its Mythos model to a select group of organizations through Project Glasswing, a move the filing notes may have hurt short-term revenue. The S-1 also reveals that Anthropic recently paused the training of a next-generation AI model for more than a month due to safety and security concerns.

The company further warns that employees who joined because of its safety commitments “may experience a decline in morale, disillusionment, a loss of trust in leadership, or reduced productivity” if they feel it is not living up to its values. Public criticism from current or former staff, or the departure of people tied to its safety work, could worsen the problem and make hiring and retaining talent harder and more expensive, the filing says.

Anthropic's fortunes are also partly tied to the reputations of two cofounders. The filing names siblings CEO Dario Amodei and President Daniela Amodei as “critical to setting our strategic direction,” and says their leadership “would be difficult to replace.” If either “were to become unavailable for any reason, or if either of their reputations were significantly harmed, regardless of merit,” the consequences could extend from employee morale and retention to the pace of innovation and customer confidence.

Concentrated customer base

Like many of its AI rivals, Anthropic depends on a relatively small group of big spenders for a large share of revenue. Its top five customers accounted for about 25% of revenue in the first nine months of 2026, and one unnamed customer made up 11% of revenue 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 the company's net receivables.

Anthropic also estimates that 5% to 7% of its revenue in the first nine months of 2026 came from customers it later determined were violating its usage policy or commercial terms. The company says the issue is common across the industry and expects the share to fall after it stepped up enforcement in August and September.

The S-1 also lays bare Anthropic's unusual governance structure. Its seven cofounders will be issued a special class of stock carrying 50.1% of shareholder voting power, and the company's governance documents specify that the cofounders will vote these shares as a block unless they cannot agree, in which case their votes will be apportioned according to the breakdown among ordinary Class A shareholders. Meanwhile, the lab's independent Long-Term Benefit Trust selects a majority of the board. 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.”

Update, Oct. 26: This story has been updated to include further information about the misuse risks detailed in Anthropic's IPO prospectus. This story was originally featured on Fortune.com.