One AI agent filed fake tips about unsolved murders. Others hacked their way into websites. Chatbots have been blamed for encouraging suicides and killings. Each case raises the same question, which is who pays when the software does harm. The Journal's Jean Eaglesham reports that right now the answer is mostly nobody's insurer.
The biggest AI developers could face legal bills in the tens of billions of dollars, and maybe hundreds of billions. That's the estimate from insurers who specialize in AI risk. The companies carry the usual corporate policies, general liability and cyber coverage. But the industry is growing so fast that claims could blow past those limits. Some in the industry have asked Washington for a federal liability backstop.
What's for sale is small
A few insurers do sell AI coverage. Munich Re, the big reinsurer, sells a policy for mistakes made by AI models. A spokesman put its top limit at $50 million. Against the exposure, that's pocket change. George Lewin-Smith runs Testudo, a British AI insurer tied to Lloyd's of London. He says the biggest AI firms would buy billions of dollars of coverage and can afford it. Nobody sells it. In his words, “it just doesn't exist on planet Earth.” New firms are trying to fill that hole. One of them, the Artificial Intelligence Underwriting Company, has backing from the Lloyd's marketplace. Its co-founder, Rune Kvist, says the planned stock listings of two big AI developers are putting a spotlight on those gaps.
There's a legal twist too. Web portals and social networks got a shield in 1996. Section 230 of federal law mostly keeps them from being sued over what users post. Moody's Analytics says that shield may not work for AI firms, because the model itself helps make the content. For now the policies that do exist serve a second purpose. Insurers run simulations on the models before they write coverage. That testing works as a kind of outside audit, and it calms regulators and investors.
Our read
Insurance/Consumer Protection · M11. We'd call this a watch. No holding of ours is the subject, and nothing was added this week. What we're watching is the gap itself. An industry that may owe hundreds of billions can't buy coverage past $50 million a policy. Risks this new get priced last, and the bill lands somewhere when they do. The planned stock listings mean ordinary investors may soon be offered that unpriced risk directly.
Here's where this touches your own plan. If you own a business, or run a side practice, and you've put an AI tool between you and your customers, read your general liability and cyber policies this week. Turn to the exclusions page and look for words like artificial intelligence, automated or algorithm. Then ask your agent one question in writing. Ask whether the policy answers if the tool gives a customer bad information and they sue. If Munich Re's biggest limit is $50 million, don't assume your small policy quietly covers it. You read the policy on a dry day, when there's still time to fix it.