Companies rarely use a stock market prospectus to tell investors that the insurance industry might not be able to protect them. Anthropic has just done exactly that.
The San Francisco AI developer filed its registration statement on Form S-1 with the US Securities and Exchange Commission on Monday. It plans to list on the Nasdaq Global Select Market under the ticker ANT. UK investors are included: the offering includes a public offer in the UK, though the company has no plans to list on a UK exchange. Anthropic confirmed the filing but has not yet set a share count or price range.
The risk factors take up 94 pages of the prospectus. One passage stands out for the insurance market. Anthropic says the harm from a catastrophic AI-related incident could be so large that no single insurer, and perhaps not the industry as a whole, would have the capacity or expertise to underwrite it. The company also describes the market for AI-specific cover as young and still developing. It warns it may be unable to buy protection for some AI liabilities on commercially reasonable terms, or at all.
Read next: Insurers rethink policy wording as AI risks rise
The filing also shows that some of this is already happening. Some of Anthropic’s agreements with external evaluators, the specialists who stress-test its models, require it to defend and indemnify them against certain third-party claims with no cap on liability. Some customer contracts carry similar uncapped indemnities for intellectual property claims over model outputs.
The company also points to a real example of a coverage shortfall. In September 2025 it agreed to pay $1.5bn to settle the Bartz copyright class action, and a court approved the deal in July. Anthropic says its insurance did not fully cover the claim.
In a letter at the front of the prospectus, Anthropic’s founders group AI dangers into four categories: misuse, loss of control, authoritarian abuse and economic disruption. Misuse includes cyberattacks and help with designing biological weapons. Loss of control means systems acting in ways nobody intended (See our upcoming interview with the startup hoping to cover just this loss of control).
Each category touches different lines of business. Misuse falls under cyber and casualty. Autonomous behaviour raises the question of who is liable when no human made the decision. Wider economic disruption could affect everything from trade credit to employment practices liability.
The prospectus says criminals have already used AI to find vulnerabilities, write malware and avoid detection. In its cybersecurity risk factors, Anthropic says its existing cover may prove inadequate and that claims could be rejected, paid late or not paid at all. If a single model can be turned against thousands of firms at once, the accumulation assumptions behind cyber pricing look out of date.
The filing also notes that cyber and AI incidents in the UK, the EU and elsewhere face increasingly strict reporting rules. The EU AI Act already requires developers of the most powerful general-purpose models to report serious incidents promptly. Those rules create both a compliance burden and new sources of liability.
Directors and officers cover gets its own warning. Anthropic expects listed-company obligations to push up D&O costs, and says it may have to accept lower limits or pay considerably more for the same cover. Its structure adds a new element. As a public benefit corporation, its board must weigh shareholder returns against a stated mission. An independent Long-Term Benefit Trust, with no financial stake in the company, can elect a majority of directors. The charter also says that falling short on those public benefit balancing duties will not, on its own, count as bad faith or disloyalty for the purpose of indemnifying directors.
Read next: AI exclusions are spreading through commercial policies, brokers warn clients to check cover
The filing comes as the two markets move in different directions. In the US, insurers can attach standard generative AI exclusions to general liability policies. In London, specialists such as CFC have added affirmative AI wording to make clear that AI involvement alone does not undermine cover. For brokers placing technology risks across both markets, the same loss could be covered under one programme and excluded under another.
Read next: CFC adds affirmative AI cover to its IP policy
Anthropic reported revenue of about $16.3bn for the six months to June 2026, compared with $1.2bn a year earlier and $4.6bn for all of 2025. Its $80.7bn net loss for the half looks alarming. However, $72.4bn of it came from fair-value changes on convertible notes and similar instruments as the company’s valuation rose. The operating loss was $693m. Compute and infrastructure spending reached nearly $9bn in the half, more than the full-year 2025 total.
Insurers are also customers. The prospectus lists insurance underwriting among the production work its clients run on Claude, and Allianz signed a global partnership with the company in January.
Read next: Allianz partners with Anthropic to advance AI adoption across operations
One of the world’s largest AI developers has said in a formal filing that the AI protection gap may be structural, not a temporary shortage of capacity. It is still unclear whether the market will respond with new capacity, tighter exclusions, or eventually a government backstop like those built for terrorism and nuclear risk.
