Banks looking to insure against AI hallucination

As banks push further into consumer-facing gen AI tools, a question is surfacing inside compliance and legal departments: Who pays when the tools get it wrong? And to solve that problem, a small cohort of insurers are betting that AI liability will become its own line of coverage. Testudo, a startup backed by Lloyd’s of London, launched what it says is the first AI liability insurance product in January, with policy limits up to $10 million and annual premiums generally in the $10,000 to $20,000 range, founder and Chief Executive George Lewin-Smith told FinAi News. The idea grew out of his time at Goldman Sachs, where he watched risk and compliance teams struggle to price the liability for roughly 400 internal AI deployments, he said. “If Goldman was struggling with this risk … everyone was going to struggle with that,” Lewin-Smith said. Testudo underwrites using historical litigation data drawn from lawsuits about gen AI use rather than focusing on model-performance audits, he said. “Litigation does correlate with the performance of an AI model,” Lewin-Smith said. However, firm-specific factors, such as industry, jurisdiction and revenue, along with which models and vendors a company uses, drive most of the exposure. The startup is an approved Lloyd’s of London’s coverholder, or company that has been authorized to enter into contracts on the insurer’s behalf, he said, adding that Testudo receives underwriting capacity and support from members of the Lloyd’s marketplace. Demand for Testudo is growing, Lewin-Smith said. “We’ve had banks come to us, say, ‘Can we have $500 million of coverage,’ and we had to say no to that,” he said, because Testudo can’t handle that scale yet. “But it’s a good indication that the market is forming,” he said. AI insurance appetite As fintechs race to develop solutions that can help safeguard AI deployments, banks are pondering if they would need such protections. “We haven’t really introduced any AI externally,” Kal Majmundar, chief technology and transformation officer at Patelco Credit Union, told FinAi News. “Even our chatbots are very prescriptive; based on the question, it has the response. You can’t get in trouble when you have a predefined answer to a question.” But Majmundar said that changes once AI moves from scripted responses to executing transactions without human review, such as moving money between accounts. FIs will evaluate the need for insurance as AI regulation starts taking shape, Lewin-Smith said. The alternatives Not every fintech executive is convinced outside insurers are the right answer. “I know the risk that I’m exposing my customers to,” Justin DiPietro, co-founder and chief strategy officer at AI-driven customer engagement service provider Glia, told FinAi News. “If you have an insurance company sitting outside the walls, they’re insuring based on risk, but … the cost of that insurance will always be higher than if I offer the terms myself because they don’t know the actual risk.” Glia provides FIs with an AI anti-hallucination guarantee for some of its products, DiPietro said, adding that FIs are very cautious about what tools they deploy because the industry is heavily regulated. Others are skeptical that a new product category is necessary. Vincent Calcagno, an executive managing director at asset servicer Ocorian, told FinAi News that carriers in investment management have largely stayed silent on AI in policy language so far — a posture he called preferable, since ambiguity tends to favor the insured. He compared the wave of AI-insurance marketing to the Y2K buildup. “Much of what might be happening could be described [as] an attempt by the insurance community to carve out a marketing angle to sell policies by creating fear/uncertainty,” he said. Register here for the FinAi Lending Summit, set for Oct. 7-8 in Las Vegas. This inaugural event will include speakers from Fifth Third and Capital One as well as a fireside chat with Piermont Bank founder and Chief Executive Wendy Cai-Lee.