Anthropic's IPO Filing Flags Insurance Gaps That Could Reach Every Company Using AI

Anthropic's IPO filing lays out a long list of risks, from intellectual property theft and security incidents to data loss, climate change, rogue agents and even an existential threat to humanity. Running through that language is a basic question: can all of the ways AI might go wrong actually be insured? Even with more cash than almost any company in America, Anthropic could not answer with confidence.

The S-1 says the company's models may behave in ways no user or developer intended, and that resulting claims could bring substantial liability. It adds that its insurance may not be sufficient, particularly when one incident or a group of incidents causes disproportionate harm. It also states that no single insurer, and possibly not the insurance industry as a whole, may have the capacity or expertise to underwrite coverage matching a catastrophic AI-related incident. The market for AI-specific risks is described as nascent, and Anthropic says it may be unable to obtain some coverage on commercially reasonable terms, or at all.

Those concerns reach well beyond chatbots that book flights or compare backyard grills. Chatbots have increasingly gone rogue, and some companies are blocking them. In extreme cases they have been accused of contributing to a death, including a case revealed this week. Alissa Lugo, a senior director analyst in Gartner's legal and compliance practice, says her research shows "death by AI" claims, meaning cases where AI played some role in a fatality, could exceed 2,000 worldwide by the end of 2026. She says more companies are adding AI exclusions and that insurers, with help from the courts, will ultimately set the parameters of coverage.

Many experts compare the situation to the early days of cyber insurance. Dennis Hendrickson of Hendrickson Insurance, an independent brokerage in Sarasota, Florida, says cyber policies currently do the most to cover AI, but only for losses such as data breaches, privacy violations, outages and deepfake wire fraud. Nearly every cyber policy excludes bodily injury, which he called a "gaping hole." Kara Higginbotham of Zurich North America agreed that a gap generally exists for bodily or physical harm from AI incidents, because many carriers are wary of writing it into policies while the scenarios remain too new to underwrite well.

Henrietta Akpata of Marist University said insurers are telling clients, "We can't cover what you can't show us," and that most companies cannot say where employees use AI tools. She noted the issue is not limited to big tech: a local business whose website chatbot gives a wrong answer, or an HR team using AI to sort job applicants, could face problems that may no longer be covered.

Experts are divided on where AI coverage is headed. Hendrickson expects financial-loss AI coverage to become routine and priced like cyber, while bodily injury will be the hard line, with tight underwriting, low sub-limits and requirements for human review before AI output reaches a patient. Some healthcare-adjacent businesses may not be able to buy it at all. Jaymin Kim, who leads emerging technologies in Marsh's cyber practice, argues most current cyber policies are sufficient. "AI is not a peril," she said, calling it a technology through which familiar harms might appear. She said Marsh is not seeing many AI claims so far and that no wholly new category of risk has emerged, though she conceded that "maybe the operative word is 'yet.'"

Vikram Desai, global cyber strategy, risk and architecture lead at Accenture, said cyber risk sat silently inside general liability for years until insurers named it, excluded it, then sold it back, and that AI is on the same curve in compressed form. He said exclusions can apply when AI sits anywhere in the causal chain, including inside a vendor's product, and that underwriters are already asking where AI is used and who oversees it.

Higginbotham said most AI-related claims today are not headline-grabbing network attacks but more routine matters, chiefly intellectual property claims against AI developers accused of scraping or training on copyrighted content, plus data misuse and privacy claims. She said pricing could rise if claim patterns shift in a sustained way, but "we are not there yet," and that Zurich does not jump to exclusions first. Hendrickson, meanwhile, expects a wave of claims that no one is prepared for, and predicts insurers, not government, will set AI standards for many businesses.