Private insurers won't cover the worst AI disasters, and many are looking to Washington

When Jeanne Eaglesham, a reporter for The Wall Street Journal who covers the insurance industry, was asked whether tech companies might turn to Washington for help with AI liability, she said there will be pressure on governments. She said many people in insurance and tech are asking, “How could we structure something that would be able to cope with that?”

By that, she said, they mean the kind of rare but catastrophic loss insurers call a tail risk. In plain terms, it is the disaster at the far end of the curve, so large that private insurers do not want to promise to pay for it. For AI, she said, that includes what she called civilization ending scenarios, which no one will underwrite.

The problem, in her account, is concentration. Risk is concentrated in only a handful of AI labs, and for labs including OpenAI and Anthropic, potential liabilities run well beyond insurance coverage, so one failure could spread widely. She offered a hypothetical example of a hack that knocked a major bank offline for an extended stretch, an event whose economic damage could be huge.

She said pressure on national governments to act is growing, and people on both the insurance and tech sides are studying the largest scale dangers from AI. She pointed to two models. After 9/11, she said, the risk of terrorism was larger than private insurers could handle, so the government stepped in to share the losses. In Britain, she said, there is a state-supported program for flood coverage.

Eaglesham said smaller AI losses could still be handled privately, noting that startup insurers are already writing AI-specific policies and large insurers are studying the market. But for now, she said, the biggest labs have liabilities well beyond their coverage, and for the systemic risks, many in insurance and tech are looking to government.