All news

28 October 2026

Anthropic's IPO filing admits some AI risks may be too big for any insurer

Anthropic warned investors that a catastrophic AI incident could exceed what insurers can cover, and insurance experts say the gap reaches well beyond AI labs.

When a company files to go public, it has to list everything that could go wrong. Anthropic, the maker of the Claude AI models, included an awkward one in its filing: some of the harm its technology could cause might be impossible to insure.

What happened

In its S-1, the document a company files with US regulators before an IPO (initial public offering, its first sale of shares to the public), Anthropic said its models may act in ways nobody intended, which could lead to large legal claims. It warned that its insurance might not cover them, and that a catastrophic AI incident could be bigger than any single insurer, or possibly the whole industry, could underwrite. It also called the market for AI-specific insurance young and said coverage may not be available on reasonable terms.

That is a striking admission from a company with more cash than almost any business in America.

Why it matters

This is not only a problem for AI labs. Businesses now use AI everywhere, from customer chatbots to sorting job applications, and insurers are still deciding which policies cover what. The research firm Gartner expects claims in which AI played some role in a death to top 2,000 worldwide by the end of 2026, and says AI exclusions are increasingly being written into coverage. Exclusions are clauses that carve a risk out of a policy.

Experts quoted by CNBC described a clear pattern:

  • Money losses are mostly handled by cyber insurance, which pays out when AI is part of a data breach, outage or deepfake fraud.
  • Bodily injury is the big gap. Nearly every cyber policy excludes it, and carriers are wary of writing it in by default.
  • Most claims so far are more mundane: intellectual property cases against AI developers over training data, plus data and privacy disputes.

Not everyone thinks AI needs its own kind of policy. Jaymin Kim of the broker Marsh argued AI is just a new route to familiar harms, saying "AI is not a peril." Others compare today to the early days of cyber insurance, when insurers first excluded the risk, then sold separate policies, and in effect wrote the security rules many companies came to follow.

What to watch

  • Whether insurers raise prices for AI-related cover if claims start to climb.
  • How courts rule on who pays when AI is somewhere in the chain of a loss.
  • How much attention investors give these risk disclosures as Anthropic's IPO moves ahead.

Based on reporting by CNBC: Anthropic warns of uninsurable AI risks. It may not be only company. Facts, not advice.

More news