Anthropic’s IPO prospectus, which the Securities and Exchange Commission made public on Monday, makes two points about advanced AI. The first is that sufficiently advanced systems could “pose catastrophic or existential risks to humanity.” The second is that this kind of self-improving AI “could adversely affect our business, financial condition, and results of operations.” Kyle Orland at Ars Technica observed, reasonably, that if the worst does happen, Anthropic’s business problems are unlikely to be high on humanity’s list of worries. Sure. But the document has one audience, which is people deciding whether to buy the stock, and for them the end of the world counts as a material risk.
It helps to know what a risk factors section is for. It isn’t really there to inform you. Its job is to make sure that if something goes wrong and the stock falls, you can’t sue the company for failing to mention it. So lawyers list everything. Jay Ritter, an IPO expert at the University of Florida’s Warrington College of Business, told Mary Cunningham of CBS News that sprawling risk disclosures in IPO filings are common. Anthropic’s is large even so. It runs to 93 pages and more than 70,000 words in a filing CBS puts at 335 pages. That is about 28% of the document, which CBS rounds up to nearly a third. You can read the S-1 itself on EDGAR if you have a free weekend.
For most companies, the disclose-everything rule produces boilerplate about competition and interest rates. Anthropic’s product is one its own makers think might be dangerous, so its boilerplate reads differently. The company says the main worry is not any current model. It is near-future models that could “substantially automate AI research and development” and, if “allowed to operate with a high degree of autonomy,” could speed up AI progress beyond the company’s ability to evaluate and safeguard each new model. The filing calls this recursive self-improvement. Under it, flaws could be passed down into later models in ways that are “difficult to detect or remediate.” This is not a far-off scenario either. As Fortune reported, the filing says “our models write a large and growing share of our code and increasingly perform our research.” It also warns that misaligned behavior during that internal work could “introduce vulnerabilities into our systems or into the training of future models.”
Then there are the risks from the models themselves. According to the filing, models could show “unexpected strategic or self-preserving behaviors,” could “resist shutdown,” and could produce “behavior resembling blackmail” aimed at preserving “the model’s continued operation or access.” They might also “possess or could develop forms of sentience or moral status.” That could bring “regulatory requirements, public pressure, or advocacy efforts.” Anthropic might also choose on its own to adopt limits that go beyond the law. Put plainly, one risk to the business model is that the product turns out to have rights.
David Hsu, who studies technology and entrepreneurship at the University of Pennsylvania’s Wharton School, told CBS by email what he thinks investors are being asked to do:
The company is asking investors to fund a race to more autonomous AI while acknowledging that capability could outpace safety. This is not an ordinary technology risk. It is a governance issue with potentially societal consequences.
Who pays
The section I find most revealing is the one about insurance, because insurance is how markets put a price on risks they can’t avoid. Anthropic says no one may be able to put a price on its biggest one: “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 or expertise to underwrite insurance commensurate with such exposure.” In other words, the whole insurance industry, taken together, might not be big enough to cover this.
Contracts are the fallback, and the filing has doubts about those as well. Customers might sue over what an AI agent did “even where the customer configured, deployed, or supervised the agent.” Legal doctrines “developed in the context of human decision-making” have not settled whether an agent’s actions create binding obligations for the user or the business that deployed it. Anthropic says it has contract terms that limit its liability, but it is unclear whether they are “enforceable or adequate.” The filing also covers minors who get around its 18-and-over rule. And it covers agents running physical equipment, where mistakes could mean “bodily injury, property damage” or “the release of hazardous substances,” with liability that is “immediate and irreversible.”
Some of the exposure comes from other companies’ products. “Even if our own models do not cause such harm, a significant safety incident involving any frontier AI could trigger broad reputational damage to the AI industry, public backlash, and sweeping regulatory responses that could adversely affect our business,” the company wrote. That means Anthropic’s shareholders also carry the risk of its competitors’ safety practices. The same reasoning applies to jobs. Anthropic worries about a “public perception” that its models cause job displacement, which could lead authorities to slow AI adoption “regardless of the actual impact of our models” on employment.
The ordinary stuff
There are normal risks too. The company’s top five customers, which it does not name, made up about a quarter of revenue in the first nine months of 2026, and CNBC quoted the filing as saying no single customer topped 10% over that period. Fortune adds that 48% of first-half revenue came through the cloud marketplaces run by Amazon, Google and Microsoft. Some local governments are considering or have approved moratoriums on new data centers. Hackers might try to get hired at Anthropic under false identities. Then there is the Pentagon dispute. The Department of War labeled Anthropic a “supply-chain risk” after the company refused to let its models be used for mass domestic surveillance or fully autonomous weapons, according to Fortune. A split D.C. Circuit panel upheld the label in September, and Anthropic has asked the full court to rehear the case.
All of this sits underneath a valuation that CBS says is expected to top $1 trillion. That number depends on the technology spreading as far and as fast as possible, while the risk section explains why that might be dangerous. In September, Anthropic CEO Dario Amodei warned that within six to 12 months AI could be able to lead a swarm of agents that might take control of the internet. We have covered the prospectus’s existential warnings and the CEO’s mixed feelings about speed before. The S-1 states both sides at once: the thing could be enormous, and that is the problem.
At least the disclosure does what a risk factor is supposed to do. If humanity does end, no one will be able to say the shareholders weren’t warned.