Insurers at Lloyd’s of London have unveiled a new insurance product designed to protect companies from financial losses caused by faulty artificial intelligence tools, including chatbot failures — a move reported by the Financial Times as a sign of the growing demand for risk mitigation in the AI era.
The policy, developed by Y Combinator-backed start-up Armilla, will cover costs incurred when AI systems underperform, such as damages awarded in lawsuits and legal fees. Several Lloyd’s syndicates are underwriting the product, targeting businesses deploying AI models in areas like customer service and decision-making where high-profile failures have already occurred.
“With companies rushing to embrace AI, missteps have already proven costly,” the Financial Times noted, referencing incidents like Virgin Money, which had to apologise after its AI chatbot reprimanded a user for saying the word “virgin.” Similarly, DPD faced backlash when its bot insulted customers, and Air Canada was forced to honour a discount fabricated by its own AI assistant — a loss that Armilla claims could have been covered under its policy had the tool’s performance fallen below expected thresholds.
Karthik Ramakrishnan, Armilla’s CEO, emphasized that the product is aimed at lowering adoption barriers for businesses hesitant about deploying AI. “We assess the AI model, get comfortable with its probability of degradation, and then compensate if the models degrade,” he explained. The insurance is activated only when an AI tool performs below a specified baseline — for example, if a chatbot that initially provided correct answers 95% of the time drops to 85%.
This targeted AI insurance fills a critical gap in the existing market. While some traditional technology errors and omissions policies offer AI-related coverage, the limits are typically low. Preet Gill, a broker at Lockton, highlighted that a general tech policy covering up to $5 million in losses might only allow $25,000 for AI-specific incidents — an inadequate figure in high-risk deployments.
Logan Payne, also a Lockton broker, added that the evolving nature of AI models — which learn and change over time — complicates coverage. Typical policies often exclude losses linked to this dynamic behavior. In contrast, Armilla’s approach focuses on quantifying degradation from initial performance levels to determine liability.
However, the policy won’t be available to all. Tom Graham of Chaucer, one of the insurers underwriting the product, stressed the importance of risk selection: “We will be selective, like any other insurance company. We won’t cover AI systems that are excessively prone to breakdown.”
As AI becomes deeply embedded in commercial operations, the arrival of dedicated coverage marks a significant evolution in both the tech and insurance industries. According to the Financial Times, this innovation may pave the way for broader AI adoption, easing fears about the potentially steep costs of machine-made mistakes.
