Insurers Set AI's Rules as Coverage Leaves Gaps

Anthropic's IPO filing warned that its insurance might not cover liabilities from AI systems, leaving the company exposed to potentially significant uninsured losses. Its models, the filing said, could behave in ways users or developers didn't intend and cause substantial liability. People harmed by those systems sit on the other side of that corporate risk calculation.

When Coverage Stops at the Data

Anthropic warned that no single insurer—and perhaps not the insurance industry as a whole—might have enough capacity or expertise to cover a catastrophic AI-related incident. It called the AI-specific insurance market “nascent and evolving” and said it might not obtain coverage for certain liabilities on commercially reasonable terms, or at all.

The filing listed intellectual-property theft, security incidents, data loss, climate change, rogue agents and an existential crisis for humanity among the risks. Existing policies, though, don't neatly cover the physical harms an AI system could help cause. Cyber insurance can address some data and security incidents, but bodily injury and physical damage remain gaps.

Alissa Lugo, a senior director analyst in Gartner's legal and compliance practice, said her research suggested “death by AI” claims could exceed 2,000 worldwide by the end of 2026. The term refers to cases in which AI played some role in a fatality. Lugo said the number was expected to grow. More companies were adding AI exclusions, and insurers, with help from courts, would determine the boundaries of coverage.

“Cyber was built for data and dollars, not for people getting hurt,” said Dennis Hendrickson of Hendrickson Insurance, an independent brokerage in Sarasota, Florida. He said cyber policies can respond to data breaches, privacy violations, system outages and deepfake wire fraud, but nearly every cyber policy excludes bodily injury. He called the missing bodily-injury or physical-damage coverage a “gaping hole.”

Everyday Uses, Unclear Protection

The exposure isn't confined to large technology firms. Henrietta Akpata, executive fellow and visiting professional lecturer of management and AI at Marist University, said a local business could face problems if its website chatbot gave a customer a wrong answer, or if a human-resources team used AI to sort job applications. “Those are everyday uses, and they may no longer be covered,” she said.

“Insurers are saying, 'We can't cover what you can't show us.' Most companies can't say where their employees are using AI tools. That's the first thing to fix,” Akpata said. Companies may not know where AI is being used, while insurers can make coverage rules depend on what those companies can document and control.

Kara Higginbotham, head of professional liability and cyber for U.S. national accounts at Zurich North America, said AI was creating exposures across insurance lines. “Every line of business is handling things a bit differently.” Zurich and some other companies have approved AI tools for employees, while some firms block them. Higginbotham said safeguards might need to become standard in policies, but a gap generally remained for bodily or physical harm linked to AI. “It is a gap because it is not being covered by default,” she said.

Hendrickson expected financial-loss AI coverage to become routine and be priced like cyber insurance, while bodily injury would remain difficult to insure. He expected tight underwriting, low sublimits and requirements for human review before AI decisions reached patients. Some healthcare-adjacent businesses, insurance experts said, might not be able to buy coverage.

The Insurers' Checklist Becomes the Rulebook

Vikram Desai, global cyber strategy, risk and architecture lead at Accenture, said the problem was “what happens when it goes wrong” and that AI had exposed how badly it fits policies companies already own. He compared the situation with early cyber insurance: “Cyber risk sat silently inside general liability for years until insurers named it, excluded it, then sold it back. AI is on the same curve, but compressed.” Desai said exclusions could apply when AI sat anywhere in the causal chain, including inside a vendor's product.

Hendrickson said insurers' checklists could set company standards before the government did. “We've seen this before with cyber insurance. First insurers stopped covering it, then they created separate policies for it, and their checklists ended up setting the rules companies followed,” he said. He added that insurers might introduce a product, receive claims and then refuse to renew policies. Independent agents, he said, often had to deal with uncovered or angry customers.

Jaymin Kim, who leads emerging technologies in Marsh's cyber practice, said Marsh wasn't seeing many AI claims and that most current cyber policies were sufficient. “AI is not a peril. It's just a technology through which existing familiar categories of harms and losses might manifest,” Kim said. She said separating AI into its own insurance category could be tedious, and that the industry hadn't yet seen a wholly new category of risk created by AI—“maybe the operative word is 'yet.'”