Anthropic is asking investors around the world, including in Australia, to back its stock market debut. In the same document, it admits that if its AI ever caused a genuine catastrophe, the insurance industry might not be able to absorb the loss.
The US developer of the Claude models filed its Form S-1 with the US Securities and Exchange Commission on Monday and plans to list on Nasdaq under the ticker ANT. According to the filing, the Australian leg of the offer will be made through a separate prospectus lodged with ASIC, although Anthropic has no plans to list on the ASX. The company has confirmed the filing but hasn’t yet set a share count or price range. All figures below are in US dollars.
Underwriters will want to read one passage in the 94-page risk factors section really closely – it flags some potentially industry shattering risks. Anthropic says the damage from a catastrophic AI-related incident could exceed what any single insurer, or possibly the market as a whole, has the capacity or the expertise to underwrite. It calls the market for AI-specific cover nascent, and says some of its AI liabilities may not be insurable on commercially reasonable terms, if they can be insured at all.
Rian Wickens, head of cyber and IT liability, professional and financial risks, at Liberty in Australia, said the market shouldn’t dismiss that as boilerplate.
“When a frontier AI developer tells regulators the insurance industry may not be able to absorb its worst-case risk, that’s more than a standard disclaimer. It’s a signal,” he said.
And he’s not alone in flagging the risks that Anthropic has highlighted. “As our understanding of the risk deepens, particularly its potential for systemic and aggregated losses, it cannot be assumed that all AI exposures will be insurable,” Andrew Hill, global FINEX AI risk strategic leader at WTW warned IB.
Read next: Clients are buying cyber insurance. That does not mean they are covered
Wickens said the filing raises three questions for underwriters right now. The first is about cover that has already been written: “Where is AI exposure already hiding in our cyber, tech E&O, PI, D&O and casualty wordings without being named?”
The prospectus suggests that’s a fair concern. Some of Anthropic’s agreements with the outside firms that test its models require it to defend and indemnify them against certain third-party claims with no cap on liability, and some customer contracts carry similar uncapped promises for intellectual property claims over model outputs. The company has also seen its own cover fall short. When it agreed in September 2025 to pay $1.5bn to settle the Bartz copyright class action in the US, its insurance didn’t pay the full amount.
Australian brokers are seeing this in claims data too. A Gallagher survey reported by Insurance Business Australia found that one in five insurance professionals had a client suffer an AI-related loss or claim in the past year, and only just over half of those were fully covered. Some insurers have started to tackle the problem directly. CFC, for example, has added affirmative AI cover across a range of its policies so that AI involvement on its own doesn’t leave a loss in limbo.
Read next: CFC adds affirmative AI cover to its IP policy
Wickens’ second question goes to accumulation: “How do we model the build-up of risk when one AI model sits underneath thousands of our insureds?”
Anthropic’s own numbers show why it matters. As of September 30, the companies paying it at least US$100,000 a year included every member of the Fortune 10, more than half of the Fortune 500 and over 30% of the Global 2000. The filing also warns that as AI systems take on more work with less human oversight, they could take harmful actions independently and at scale, with nobody stepping in to stop them.
There’s a supply chain issue as well. Anthropic leans on Amazon, Google and Microsoft for much of the computing power behind its models, and says its own insurance may not cover the business interruption losses that a serious outage would cause. The filing also notes that it has signed leases for data centre capacity in Australia.
Reinsurers are already reorganising around this kind of risk. Gallagher Re recently set up a digital risk practice after its own research identified AI model failures as a source of aggregation across ceded portfolios.
Read next: Gallagher Re creates digital risk unit as AI reshapes accumulation
The filing is arriving as Australia’s prudential regulator is also flagging risks. In an April 30 letter to all regulated entities, APRA said AI adoption was materially changing the cyber threat landscape, and that governance and risk practices weren’t keeping pace. Anthropic’s prospectus makes a similar point, noting that criminals are already using AI to find vulnerabilities, write malware and get past detection tools.
For insurers using AI in their own operations, and Anthropic lists insurance underwriting among the production work customers run on Claude, the question also falls under APRA’s CPS 230 operational risk standard. The standard came into force on July 1, 2025, and the transition period for existing service provider contracts ended this July. Under it, providers of core technology services are treated as material service providers unless an entity can justify otherwise, so an AI model that underpins underwriting or claims is likely to need managing on that basis.
Read next: Revealed – best cyber insurance providers in Australia for 2026
Wickens’ final question is the one the market has struggled with for every new systemic risk: “If the worst-case losses can’t be insured, what’s the backstop?”
Australia has an existing model for this kind of problem. Terrorism risk sits partly with a government reinsurance pool run by the Australian Reinsurance Pool Corporation, created after private capacity pulled back. Nothing comparable exists for AI, and Anthropic’s filing doesn’t suggest one.
Wickens thinks the pattern is a familiar one. “We’ve been here before on silent risk in insurance policies with terrorism, pandemic and cyber,” he said. “AI won’t be any different.”
Anthropic’s governance adds a further wrinkle for D&O underwriters. As a public benefit corporation, its board must balance shareholder returns against a stated mission, and an independent Long-Term Benefit Trust with no financial stake in the business can elect a majority of directors. The company also expects listing to push up its D&O costs, and says it may have to accept lower limits.
Insurers, meanwhile, are among its customers. Allianz signed a global partnership with Anthropic in January to use Claude across its operations.
Not everyone in the industry, however, shares the AIpocalypse views. “I think it's just going to become part of our lives, and therefore we need to learn how to live with that,” Daniel Elechiguerra Batlle , a cofounder of specialist AI agent insurer Harbor told Insurance Business. “I think guardrails and the controls are going to improve, so it's not something that is going to happen happen in the same magnitude that we're seeing here initially.”
Read next: Allianz partners with Anthropic to advance AI adoption across operations
Do the client’s cyber, tech E&O, PI, D&O and casualty wordings mention AI at all, and if not, which policy would respond to an AI-related loss?
