Anthropic tells investors a catastrophic AI incident could exceed what insurers can cover

In the risk factors of its Form S-1, Anthropic, the developer of the Claude models, says its insurance may not be enough to cover liabilities from harm linked to its models, in particular when one incident or a group of incidents causes very large harm. Given the possible scale of a catastrophic AI-related incident, it says that no single insurer, and possibly not the insurance industry as a whole, may have the capacity or expertise to write matching cover.

The filing describes the insurance market for AI-specific risks as "nascent and evolving". Anthropic says it may not be able to obtain cover for some AI-related liabilities on commercially reasonable terms, or at all, which could leave it with potentially significant uninsured losses. It gives a similar warning about cyber and data liability cover, and adds that claims it submits may not be covered, paid on time or paid at all. On litigation, it says insurance may not fully cover claims, and cites the Bartz copyright case as an example.

Anthropic has also taken on some liability by contract. To support independent testing of its frontier models, certain agreements with external evaluators and testing vendors require it to defend and indemnify them, with no cap, against some third-party claims, including allegations that its models or their outputs infringe intellectual property. Separately, the filing says the legal rules on liability for actions that AI systems take on their own, including its agentic products, are unsettled.

Morningstar DBRS said on 27 October that the filing puts the AI insurance protection gap in focus, Reinsurance News reported. The rating agency said catastrophic AI liabilities could exceed the capacity of the insurance industry, and that uninsured liabilities remain a possible claim on the developer's capital and liquidity. It warned that policies without specific AI cover could carry "silent AI exposure", which makes pricing and reserves uncertain, and that one model failure could cause claims from many policyholders at once, because diversification works less well when clients use the same technology.

According to the same report, Anthropic expects its D&O insurance costs to rise as a public company, possibly with lower limits. The listing could come as soon as mid-November.