Anthropic IPO filing warns AI risks may be uninsurable

A massive risk disclosure

Anthropic is preparing for a potential stock market debut that could value the company at up to US$2 trillion. As part of this process, the company filed a prospectus with US securities regulators that includes 93 pages dedicated to risk factors. This is a substantial amount of documentation compared to other high-risk enterprises like SpaceX, which included only 38 pages of risks in a similar filing.

The company categorizes these potential dangers into four specific buckets. These include misuse of the technology for bioweapons or cyberattacks, the potential loss of human control over systems, authoritarian abuse, and general economic disruption. The filing even goes as far as flagging existential risks to humanity as a possibility.

Legal experts note that prospectus risk sections are primarily written to shield directors from future lawsuits. Therefore, the length of these sections serves as a clear indicator of what a company expects it might be sued over. By disclosing these issues now, Anthropic is establishing a public record of the dangers associated with its frontier models.

The limits of insurance

One of the most significant warnings in the filing concerns the ability of the insurance market to handle AI-related liabilities. Anthropic states that no single insurer, and potentially the insurance industry as a whole, may have the capacity or expertise to underwrite risks commensurate with these exposures. This suggests that the financial fallout from a major AI incident could exceed what current insurance products are designed to cover.

The insurance market is already showing signs of hesitation regarding AI. While Gartner estimates that 'death by AI' claims could exceed 2,000 worldwide by the end of 2026, most current cyber policies almost entirely exclude bodily injury. Carriers like Zurich North America have expressed wariness about providing coverage for such specific types of harm.

The difficulty in providing coverage is compounded by a lack of visibility into how companies actually use these tools. Many organizations cannot identify where their staff are using AI, which makes it nearly impossible for insurers to accurately price risk. Without clear data on usage, many firms may find themselves without coverage if an AI-driven incident occurs.

Contagion and regulation

Anthropic also warns about the concept of 'contagion' within the AI industry. The company notes that even if its own models do not cause harm, a significant safety incident involving any other frontier AI could trigger broad reputational damage and public backlash. This means that a disaster at a competitor's company could lead to sweeping regulatory responses that affect everyone in the space.

The company also acknowledges that its own lobbying efforts for AI rules could lead to complications. It warns that these actions could subject the firm to heightened scrutiny, political backlash, or disproportionate regulatory action. This is a practical concern, as the company has already faced friction with the Trump administration over the Pentagon's use of its models.

The US Defence Department has already labeled Anthropic as a national security risk and barred it as a supplier. These real-world tensions highlight how quickly AI technology is moving from a laboratory setting into high-stakes government and commercial environments. The company's disclosures suggest that the regulatory landscape will be shaped by these high-profile conflicts.

Impact on business owners

For business owners, these disclosures mean that AI should be treated as a material operational risk rather than a simple productivity tool. Because the company's own developers cannot fully guarantee that their systems will not cause harm, the liability for using them falls on the businesses that deploy them. This is particularly relevant for directors who owe a duty of care regarding material risks under corporate laws.

Firms that lack documented AI governance frameworks may already be facing coverage denials at insurance renewals. The filing makes it harder for companies to wave away vendor dependence or liability. If a business uses AI to perform core functions, it must be prepared to handle the legal and financial consequences if those systems fail or behave unexpectedly.

To manage this, businesses should map out exactly where AI is being used across their operations, including tools adopted quietly by employees. They should also review vendor terms for liability caps to understand who carries the loss in a disaster. Finally, companies should ask their brokers in writing whether their current cyber and D&O policies respond to harm caused by AI systems.