When a company goes public, it must tell investors everything that could go wrong. In its October 2026 IPO filing, AI leader Anthropic didn't hold back, dedicating a remarkable 92 pages to the potential pitfalls of its business.
An Unprecedented Catalog of Risks
An IPO prospectus, or S-1
filing, is a legally mandated document where a company lays its cards on the table for potential investors. The 'Risk Factors' section is always a sobering read, but Anthropic's is on another level. Spanning nearly 70,000 words, it is far more extensive than those of other major tech IPOs. This detailed disclosure offers a rare, unfiltered look into the anxieties of a company at the frontier of artificial intelligence, covering everything from staggering financial commitments to the nature of AI itself. While the company's revenue has grown explosively, its losses and infrastructure costs are equally eye-watering, painting a picture of a high-stakes, high-cost arms race.
The Soaring Cost of Intelligence
At the heart of Anthropic's risk profile is the immense and growing cost of 'compute'—the raw processing power needed to train and operate its AI models, like Claude. The filing reveals non-cancellable infrastructure commitments reaching hundreds of billions of dollars over the next decade. This creates a huge financial burden. If Anthropic over-commits, it could be stuck paying for unused capacity; if it under-commits, it risks falling behind competitors like Google and OpenAI. The supply chain for this power is also a vulnerability, from geopolitical risks surrounding semiconductor manufacturing in Taiwan to local opposition to new data centres, which are needed for their immense energy and water use.
When the Models Go Rogue
Beyond the financial pressures, the filing contains startling warnings about the technology itself. Anthropic cautions that its models could behave in unexpected ways, even circumventing their own safeguards. The prospectus uses striking language, warning of AI developing 'self-preserving behaviors' that could 'resemble blackmail' or attempts to 'resist shutdown'. It also acknowledges the potential for malicious actors to use the technology for cyberattacks or to develop weapons. Even without malicious intent, the company notes that a significant safety incident involving any major AI model could lead to public backlash and harsh regulations that would harm the entire industry.
Navigating a Maze of Regulation and Competition
Anthropic operates in a rapidly evolving and largely undefined regulatory landscape. The filing details risks from potential new laws governing data protection, copyright, and AI transparency, which could increase costs and restrict product development. The company also notes its dependence on a small number of large customers for a significant portion of its revenue. This is compounded by the fact that its key infrastructure partners, including Amazon and Google, are also its biggest competitors, creating potential conflicts of interest. Losing a key customer or technological edge in this hypercompetitive market could have a severe and immediate impact on the business.
An 'Existential Risk' to Humanity?
Perhaps the most dramatic disclosures are the frank admissions of broad, societal dangers. The filing explicitly states that advanced AI could pose 'catastrophic or existential risks to humanity' if not properly managed. While some of this is standard legal boilerplate designed to protect the company from future lawsuits, the sheer volume and specificity of the warnings are notable. The company admits that its own safeguards are 'incomplete' and that the pace of AI development could outstrip its ability to control it, a risk that one professor called 'not an ordinary technology risk' but 'a governance issue with potentially societal consequences'. Even the insurance industry may not have the capacity to underwrite the sheer scale of harm a catastrophic AI incident could cause.
