Key points
- Anthropic warns an AI catastrophe could exceed insurers' capacity
- Some Anthropic tester contracts leave certain claims uncapped
- Anthropic owed $1.2 billion of its book settlement as of June 30
Anthropic says the insurance industry may not have the capacity or expertise to cover losses from a catastrophic AI incident. In its IPO filing, made public October 26, the company warns that the exposure could exceed what a single insurer, or potentially the industry as a whole, can cover.
The filing also describes a separate source of potential liability. Under some agreements with outside testers, Anthropic has agreed to defend and pay certain third-party claims without a cap, including claims that its models or their outputs infringe someone else's intellectual property during authorized testing.
The insurance warning appears in a section on harm that could be caused by Anthropic's models. "Our insurance policies also may not provide sufficient coverage to adequately mitigate the potential liabilities we face, especially where any one incident, or a group of incidents, causes disproportionate harm," the filing says. It calls the market for AI-specific insurance "nascent and evolving." Anthropic may not be able to buy coverage for some AI-related liabilities on commercially reasonable terms, "or at all, leaving us exposed to potentially significant uninsured losses."
The filing lists the claims Anthropic could face
If its models are involved in causing significant harm, Anthropic says it could face personal injury or negligence claims, regulatory enforcement actions, and criminal investigations. Its terms of service disclaim warranties and limit its liability. The company says those protections may not be upheld, or may not apply to certain claims.
"Because our models may behave in ways that no user or developer intended, claims arising from such actions could result in substantial liability," the filing says. A harmful incident, "regardless of fault," could also damage Anthropic's reputation as a safety-focused AI company and cost it customers.
Some tester contracts lift the cap for certain claims
In some cases, Anthropic has taken on extra liability by contract. To make independent testing of its frontier models possible, it agreed to cover certain outside evaluators and testing vendors for losses from that testing. Those losses include security incidents, data loss, and data breaches.
The agreements generally limit what Anthropic owes, according to the filing. Some, though, require it to defend the evaluator against certain third-party claims and pay for them "without a cap on our liability." They include claims that Anthropic's models or their outputs, as used in the evaluator's authorized testing, infringe someone else's intellectual property.
The filing doesn't say which evaluators have the uncapped terms. Elsewhere, it says government AI evaluation bodies in the US and the UK get early access to its recent frontier models, but it doesn't connect them to those arrangements. Separately, Anthropic said earlier this month that it had given six researchers from the nonprofit METR access to its internal systems to assess its AI risks.
The book settlement is the filing's own example
Anthropic points to its own copyright case. In September 2025, it agreed to pay $1.5 billion to settle Bartz v. Anthropic, a class action brought by authors. The district court approved the settlement in July 2026. "Moreover, as in Bartz, insurance may not fully cover such claims," the filing says in its section on litigation.
Anthropic recorded the full $1.5 billion as an expense in 2025, equal to 33% of that year's $4.59 billion in revenue. It paid $300 million in 2025. It still carried $1.2 billion of the settlement as a current liability on June 30, 2026, according to notes to its financial statements.
Music publishers, including Concord Music Group and BMG Rights Management, have also sued Anthropic over song lyrics.
Cyber coverage carries the same warning
Anthropic says it currently has insurance that may cover some costs of security breaches and other incidents. It may still be unable to get coverage for certain cybersecurity or data-related liabilities on commercially reasonable terms, or at all. There's also no guarantee its claims will be "covered, paid in a timely manner, or paid at all" by its insurers, the filing says.
Anthropic also expects the rules for public companies to make director and officer liability insurance more expensive. It says it may have to accept lower policy limits or pay much more for the same coverage.
Insurers are still deciding how to cover AI
A gap generally still exists in coverage for bodily or physical harm from an AI-related incident, Kara Higginbotham of Zurich North America told CNBC. She heads professional liability and cyber for its US national accounts. "It is a gap because it is not being covered by default," she said.
Gartner analyst Alissa Lugo said her research shows "death by AI" claims could exceed 2,000 worldwide by the end of 2026. Gartner counts any case in which AI played some role in a death, which doesn't mean AI caused it. More companies are building in AI exclusions, she said.
Anthropic plans to list on the Nasdaq Global Select Market under the ticker ANT. Its public S-1 doesn't yet include a price range or the number of shares it will sell.
