If Anthropic can’t insure its own AI, your board can’t ignore it

Look past the doom headlines and Anthropic’s IPO prospectus says something more useful to business owners than “AI could end humanity”. The company behind Claude is preparing a listing that could value it at up to US$2 trillion. It has now told US securities regulators, with legal liability attached to every word, that the worst outcomes of its technology may sit beyond the reach of the insurance market.

That is not ethics theatre. It is a pricing signal, and any New Zealand business wiring these tools into its operations should treat it as one.

Ninety-three pages of things that could go wrong

Of the filing’s 335 pages, 93 are devoted to risk factors. SpaceX, hardly a low-risk enterprise, got by with 38 pages of risk in a filing of more than 250. Prospectus risk sections are written by lawyers to shield directors from future lawsuits, so their length is a fair measure of what a company expects it could be sued over.

Anthropic sorts the dangers into four buckets: misuse such as bioweapons and cyberattacks, loss of human control, authoritarian abuse and economic disruption, according to RNZ’s reading of the filing. It goes as far as flagging existential risk to humanity.

The commercially sharpest line is about contagion. Anthropic warns that “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“. In plain terms, one competitor’s disaster becomes everyone’s regulatory problem. It also concedes its own lobbying on AI rules “could subject us to heightened scrutiny, political backlash, or disproportionate regulatory action”. That is not hypothetical. The company has already clashed with the Trump administration over Pentagon use of its models, and the US Defence Department has labelled it a national security risk and barred it as a supplier, the Herald reports.

Growth this fast buys a lot of exposure

The numbers explain why the lawyers were busy. April-June 2026 sales hit US$11.6 billion, up from US$787 million a year earlier, delivering a first operating profit of roughly US$1 billion. Yet the company still booked an overall net loss exceeding US$80 billion in the first half, driven by computing costs and valuation-linked accounting charges.

A business growing at that pace, deployed by that many customers, has a liability surface that grows just as fast. Every new enterprise contract is another place where something can go wrong.

The insurers are already blinking

Here is the part New Zealand boards should pin to the wall. Anthropic warns that “no single insurer, and potentially the insurance industry in the aggregate, may have sufficient capacity or expertise to underwrite insurance commensurate with such exposure”.

The market is not ready. Gartner estimates “death by AI” claims could exceed 2,000 worldwide by the end of 2026, yet current cyber policies almost all exclude bodily injury, and carriers such as Zurich North America remain wary of writing that cover. Worse, companies often cannot even identify where staff are using AI tools, which makes underwriting guesswork.

Locked out of the upside, exposed to the downside

The New Zealand twist is irritating. The prospectus is not registered under the Financial Markets Conduct Act, so ordinary Kiwi retail investors cannot buy in while Australians reportedly can. Local firms get none of the equity upside but every bit of the operational risk the company has just disclosed.

That risk lands on directors. Under the Companies Act 1993 they owe a duty of care on material operational risks, and as Insurance Business Magazine puts it, “an AI system that its own developer cannot fully guarantee will not cause harm is a material operational risk”. The same analysis warns that firms without documented AI governance frameworks are now facing coverage denials at renewal. Commentators have also argued the filing makes vendor dependence and liability far harder to wave away for Kiwi firms weighing AI adoption.

What to do before the next renewal

None of this is an argument against using AI. The productivity case is real, and businesses that sit it out will lose ground. It is an argument for treating AI like any other material supplier risk rather than a free productivity hack.

That means mapping where AI is actually used across the business, including the tools staff have quietly adopted. It means reading vendor terms for liability caps, because those caps tell you who carries the loss. It means asking your broker, in writing, whether your cyber and D&O policies respond to AI-caused harm. And it means documenting a governance framework before an underwriter asks for one.

Anthropic lists on Wall Street within weeks, and its disclosures will become the benchmark every other AI vendor is measured against. The smart money is no longer asking whether AI risk is real. It is asking who ends up holding it. Boards that cannot answer that question for their own business should expect their insurer to start asking it for them.

Sources

• RNZ: Anthropic set to make stock market debut as questions swirl about safety of AI (2026-10-27)
• RNZ: Kiwis shut out of Anthropic IPO, while Aussies get a seat at the table (2026-10-28)
• 1News: Anthropic warns investors of blowback risks from runaway AI (2026-10-31)
• NZ Herald: Anthropic unveils IPO filing, showing soaring sales and AI risks (2026-10-26)
• CNBC: Anthropic warns of uninsurable AI risks. It may not be only company (2026-10-28)
• Insurance Business Magazine: Anthropic says its AI could end humanity – putting pressure on NZ boards (2026-10-27)