Insurance Faces a New Risk Problem: AI Moves Faster Than Historical Data
For decades, insurers priced uncertainty by looking backward. Historical claims data told them what would happen next. That model is breaking down as artificial intelligence systems evolve faster than the patterns that once predicted them.
The result is a fundamental shift in how risk works. It's no longer static enough to classify and file away. Risk now changes speed, arrives wrapped in ambiguity, and operates at scales that traditional systems weren't built to handle. Underwriters must price policies for autonomous AI agents that don't yet have claims histories. They must account for concentration risk when three foundational model providers power thousands of businesses. They must do this while their own underwriting systems are being automated.
Three insurance startups offer a window into how the industry is adapting.
Risk No Longer Stays Still
Alexandre Musy, cofounder of Huscarl, discovered early that companies fundamentally misunderstand their own risk exposure. Some requested insurance coverage five times their actual maximum exposure. Others had no idea what their actual risk looked like.
The problem wasn't stupidity. It was quantification. Companies lacked the tools to measure what they were actually exposed to.
Huscarl works with corporations generating more than $50 million in revenue to optimize expensive insurance programs. The company relies on actuarial science rather than intuition to determine what risks exist and how to price them. Musy says optimization often cuts premiums by 30 percent.
AI risk creates a specific problem: there's no statistical history to draw from. "The issue is that it's an unknown risk in the sense that we don't really know what frequency and severity it happens at," Musy said. That's similar to where cyber insurance stood in its early days-insuring something without a claims baseline to work from.
Insuring the AI Systems Themselves
A new category of risk has emerged: losses caused by AI agents themselves. Klaimee, founded by Ines Boutemadja and Julien Catonnet, builds insurance products for companies deploying autonomous systems in healthcare, legal services, and financial lending.
These companies hit a wall during procurement. Cyber insurance didn't cover what they needed. Klaimee underwrites the consequences when an attack against an agent succeeds or when the agent fails outright. "It covers AI agents just like professional liability would cover a consultant or service provider," Boutemadja said.
The speed of attack and defense has accelerated on both sides. Developers can build autonomous systems and mock attacks in days. Defenders respond faster too. But the deeper concern is concentration.
Three dominant model providers power most of the AI ecosystem. A failure, exploit, or hallucination affecting one of them could cascade across thousands of businesses simultaneously. In traditional insurance terms, that's catastrophic tail risk-many losses from a single event.
When AI Becomes the Underwriter
While Klaimee insures AI systems, Pibit is using AI to transform how insurers themselves work. The company's CURE platform automates the time-consuming process of evaluating commercial insurance submissions.
Akash Agarwal, Pibit's founder, identifies two blind spots in how organizations adopt AI for underwriting: governance and data provenance. "Many organizations are treating AI adoption primarily as a workflow acceleration initiative instead of a risk governance challenge," Agarwal said. Investment flows into models and automation. Explainability, auditability, and monitoring for model drift get less attention.
That matters because AI systems inherit the assumptions and biases embedded in their training data. If historical underwriting decisions contained structural bias, scaling those decisions through AI only compounds the problem.
For commercial policies involving millions in exposure-a trucking fleet with 5,000 vehicles, a factory with thousands of workers-evaluation accuracy is critical. "There's a very thin line between being technologically savvy and blinded by technology," said Prakar Mohan, head of marketing at Pibit.
In the post-AI economy, risk exists in online behavior, autonomous systems, and rapidly evolving infrastructure that insurers struggle to fully understand. The future of insurance may depend less on predicting certainty and more on adapting continuously to uncertainty itself.
Learn more about AI for Insurance and how Generative AI and LLM are reshaping the industry.
Your membership also unlocks:
