When Jeff Bezos declared in Turin this month that the world is living through a “good kind of bubble,” it struck a chord - and a nerve - across the insurance world.
For investors, the idea that some bubbles might be useful is comforting. For underwriters, it is less so. The insurance industry, long accustomed to pricing human error and corporate hubris, has seen this film before: new technology, soaring valuations, and a collective conviction that “this time it’s different.” It rarely is.
Bezos, speaking at Italian Tech Week, cast the current wave of artificial intelligence investment as a positive mania - “industrial rather than financial,” he said - comparing it to the fibre-optic buildout of the late 1990s and the biotech frenzy before it. Those bubbles, he argued, left behind networks and medicines that changed the world. “The ones that are industrial are not nearly as bad,” he said. “They can even be good... The benefits to society from AI are going to be gigantic.”
Few in the insurance sector doubt AI’s potential to reshape their own industry - from claims triage to underwriting, risk scoring, and pricing. But when bankers and regulators begin warning of speculative excess, insurers instinctively reach for the exclusion clause.
Across the market, the tone has turned distinctly cautious. Specialty carriers such as Beazley have begun warning against “AI-washing” and untested deployments, urging clients to adopt the technology only where controls and liability trails are clear. CFC, the London-based cyber and tech MGA, recently disclosed that most businesses using AI remain uncertain whether their policies would respond to an AI-driven loss at all.
Reinsurers, too, are uneasy. The Swiss Re Institute has flagged the risk of “silent AI exposure” creeping across multiple lines - from professional indemnity and cyber to product liability - where policy language lags behind technology. Some fear a replay of the early cyber era, when claims emerged years before the market had defined what it had actually insured.
Even the biggest AI developers are struggling to find cover. Reports in the Financial Times this autumn suggested that OpenAI and Anthropic have had to assemble bespoke, partially self-insured programmes, with capacity from only a handful of global markets. The sums involved - billions in potential legal exposure - illustrate just how uncharted the risk frontier has become.
For insurers, the worry is not only whether AI will underdeliver, but how badly. A sharp correction in technology valuations would hit the industry on two fronts: first, through balance sheets, where many life and property-casualty firms hold equity stakes or credit exposure to tech issuers; and second, through liability lines, if disappointed investors, partners, or consumers seek redress.
Insurers remember well the cascade of litigation that followed the dot-com bust - shareholder suits, misrepresentation claims, and D&O losses that ran into the billions.
Today’s interlocking web of AI investments, cross-holdings, and vendor financing has an unsettling resemblance. As one senior London market executive put it privately this week: “We’re insuring the plumbing of the bubble, not just the bubbles themselves.”
Should the air come out suddenly, expect losses to filter through errors and omissions, directors’ and officers’, and cyber policies alike. The complexity of AI’s supply chain means fault may prove difficult to assign.
Bezos’s optimism - that even failed bets on AI will leave behind valuable digital infrastructure - may yet prove true. History suggests that infrastructure bubbles can yield lasting dividends: the canals of the 18th century, the railways of the 19th, the fibre networks of the 1990s. Each left behind assets that later generations exploited at lower cost.
But such booms also left insurance markets nursing losses. The railway mania bankrupted dozens of 19th-century underwriters. The telecom crash of 2001 spurred years of re-pricing in directors’ and professional indemnity cover. Even “good” bubbles are rarely good for insurers in the short term.
As the Bank of England and IMF warn of stretched valuations and rising systemic leverage, the industry is again being asked to judge how much risk is truly transferable.
In truth, the insurance market is approaching AI with characteristic conservatism: tightening wordings, limiting aggregates, and favoring data-led firms over speculative ones. As Goldman Sachs chief executive David Solomon observed in Turin, “There will be a lot of capital that was deployed that didn’t deliver returns... It’s not different this time.”
For insurers, the task is not to call the top, but to survive it. That means steering clear of unquantifiable exposures, watching portfolio concentration in AI-heavy assets, and preparing for claims that will test the boundaries of cyber and liability wordings alike.
Bezos may yet be right: AI could prove a “good” bubble, an exuberant phase that builds tomorrow’s infrastructure. But if history is any guide, it is insurers who will be left sweeping up the claims when the excitement subsides.
