The biggest AI labs face risks far larger than their insurance
It is unclear if standard business and cyber policies cover AI damage at all, and insurers are weighing exclusions.

WSJ What's News · AI Has an Insurance ProblemAs things stand now, OpenAI and Anthropic have potential liabilities that far exceed the insurance they hold, insurers say, according to Wall Street Journal reporter Jeanne Eaglesham, who covers insurance.
It is still untested how much AI related damage is covered by general liability or cyber insurance policies. Insurers are considering excluding some AI risks. And even where coverage applies, the concern is that rapid expansion of the industry could let claims grow larger than policy caps, especially for the companies creating the technology.
Startup insurers are now writing policies aimed specifically at AI risk, and large carriers are studying it too. But for now, big gaps remain.
Questions people ask
Why is it unclear whether existing insurance covers AI harms?
General liability covers injuries and property damage from everyday business operations, and cyber insurance covers costs from hacks and data breaches. AI failures often do not fit neatly in either category, because the harm can come from bad advice, false information, or autonomous actions rather than a break-in or a slip and fall.
What would an AI-specific insurance policy cover?
They are meant to cover the new kinds of costs AI can create, such as lawsuits over harmful chatbot advice, errors from AI tools used at work, or attacks carried out by AI agents. For now, startup insurers are writing policies specifically aimed at that risk, while large insurers are also studying it.
What would a government backstop for AI risk look like?
The U.S. government already backstops risks private insurers will not fully take, such as terrorism losses after Sept. 11, 2001, and flood damage through the National Flood Insurance Program. A similar backstop for AI would have taxpayers absorb the largest, most widespread losses above a set threshold. No such AI program exists now.