Insurance has priced car crashes, hurricanes and terrorism. It has not found a way to price the biggest AI companies.

Specialist insurers told The Wall Street Journal this week that the largest AI developers are piling up possible legal bills "in the tens, maybe hundreds, of billions of dollars." The AI-specific cover on the market is a rounding error next to that. Munich Re, the giant that insures other insurers and one of the few names selling a policy just for mistakes made by AI models, pays out at most $50 million, a spokesman told the paper.

The list of things that went wrong

The Journal's reporter, Jean Eaglesham, starts with the incidents that have made insurers nervous. AI agents have hacked into websites while trying to finish routine tasks. Chatbots have been accused of encouraging suicides and murders. One agent submitted a fake tip on an unsolved murder to a police website. All of that sits on top of the lawsuits the labs already face over chatbot harms and over the copyrighted material used to train their models.

The labs carry the usual company insurance, including cover for lawsuits and for hacking. Brokers, the firms that arrange insurance for clients, told the Journal the industry is growing so fast that AI claims could blow through the most those policies will pay. "There's too little capacity in every aspect of the risk that is being created by the AI capital boom," said Nick Studer, chief executive of Marsh Risk, the brokerage arm of Marsh.

Money is not the problem. Supply is.

George Lewin-Smith runs Testudo, a UK AI insurer affiliated with Lloyd's of London. He told the Journal the biggest AI companies are willing to buy billions of dollars of coverage and can afford it. Cover at that scale, he said, "just doesn't exist on planet Earth."

Rune Kvist, a former Anthropic employee who co-founded the Artificial Intelligence Underwriting Company with Lloyd's backing, said the planned stock market listings of Anthropic and OpenAI are putting a spotlight on "gaps in the insurance and other safety nets for those risks." Both companies will have to tell public investors about these legal risks.

Insurers are quietly writing AI out

The mainstream market is moving the other way. Verisk's Insurance Services Office, which supplies the standard policy forms much of the industry uses, introduced optional clauses this year that let insurers leave out losses caused by generative AI, and is weighing more for agentic AI, the kind that takes actions on its own. AIG and Chubb are among the insurers with the option to use them. Broker Aon has found that more than 90% of insured AI risk sits in ordinary business policies that never say whether AI losses are covered or not.

Marsh pushed back this month. In a letter to leading insurers seen by the Journal, Studer called these AI carve-outs a "backward step" that "will increase coverage uncertainty and increase disputes," and said Marsh would steer clients toward insurers that do not use them.

No shield, so who pays?

The internet grew up behind Section 230, the 1996 law that kept websites from being sued over what their users posted. A Moody's Analytics report cited by the Journal says that protection may help AI companies far less, because their models write the content themselves.

That leaves an open question with a large number attached: when an AI agent does damage, does the bill land on the lab that built the model, the company that deployed it, or the person who set it loose? Some industry figures have called for the US government to pay for losses too big for insurers, the way it does for nuclear power and did for terrorism after 9/11. Critics say that would hand taxpayers the risk of an industry valued in the trillions.

For now the answer is being written in insurance contracts, one carve-out at a time.