As our data is fed into more large language models, the dangers of what could go wrong can easily accumulate. AI isn’t just changing the scale of risk, but the nature of how risk is best understood.
The result is a post-AI world where risk is no longer just about prediction but about speed, ambiguity and scale. From underwriting massive commercial policies to insuring autonomous AI agents, insurers are now confronting risks that are partially unknowable and moving faster than traditional systems were designed to handle.
For decades, insurers have depended on historical data to price uncertainty. AI disrupts that model because it creates systems that evolve faster than historical patterns can keep up with. So to understand what risk looks like in a post-AI reality, we asked 3 startups disrupting in insurance about their predictions on risk.
1. The End of Static Risk
In a post-AI world, risk is no longer static enough to simply classify and must constantly be recalculated. For Alexandre Musy, cofounder of Huscarl, the traditional insurance industry often misunderstood risk long before AI entered the picture. Before launching the company, Musy worked at Descartes Underwriting and studied German philosophy, a background that shaped how he thought about uncertainty and systems.
What Musy discovered early on was that many companies fundamentally misunderstood their own exposure to risk. “Something that we realised very quickly is that there were a lot of insurance submissions that were sent by our prospective customers which did not make a lot of sense from a risk perspective.” Some companies with maximum exposures of $10 million were requesting coverage five times higher. “There was really a disconnect between risk and what people were asking in terms of coverage.”
That disconnect became a problem of quantification. Rather than relying on intuition or broad categories, Huscarl leaned heavily on actuarial science to determine what risks actually existed and how they should be priced.
The company now works primarily with corporations generating more than $50 million in revenue, helping them optimize expensive insurance programs. “They’re often paying several hundreds of thousands of dollars every year on insurance premiums or even sometimes millions,” says Musy. “So if you’re able to optimize that sum and make it more efficient, you can easily cut it by 30%.”
Musy compares today’s AI exposure landscape to the early days of cyber insurance. “The issue with AI risk exposures is that you don’t have any statistical data on them. It’s very hard to put numbers on it. The main issue is that it’s an unknown risk in the sense that we don’t really know what frequency and severity it happens at.”
2. Insuring Autonomous Agents
The risks posed by AI compounds when one realizes that AI assets too are now within the purview of insurance. A new crop of companies are now providing insurance for AI agents to ensure losses brought about by them are also covered.
One such startup is Klaimee. Founded by Ines Boutemadja and her husband Julien Catonnet, Klaimee focuses on insurance products designed specifically for AI agents. The startup works with AI vendors building autonomous systems that perform tasks traditionally handled by humans.“These companies used to face a wall at procurement because people would ask them to buy cyber insurance,” Boutemadja told Startup Beat, noting how Klaimee underwrites the consequences if an attack against an agent succeeds or when the agent fails. “It covers AI agents just like a professional liability would cover a consultant or service provider.” Increasingly, voice agents and autonomous systems are carrying out sensitive tasks, in healthcare, legal services, and autonomous financial services such as lending agents making credit calls.
The scale of AI also changes the scale of potential attacks. “People are able to build agentic entities really fast and mock up an attack in a much more efficient way now,” says Boutemadja. Yet she also notes that defensive responses are becoming faster as well. “The truth is the response is also much more efficient. So you have both sides of the equation coming into play.”
The deeper concern, however, is concentration risk. Despite the explosion of AI applications, much of the ecosystem depends on only a handful of foundational models. “It’s like about 3 out there,” Boutemadja says, referring to dominant AI model providers. “Which means that insuring AI agents that rely on those has a catastrophic tail.”
In traditional insurance terms, catastrophic risk refers to the possibility of many losses occurring simultaneously from a single event. In the AI era, a failure, exploit or hallucination affecting one dominant model provider could cascade across thousands of businesses at once.
3. When AI Becomes the Underwriter
While companies like Klaimee are insuring AI systems, Pibit is focused on using AI to transform how insurers themselves operate.
Founded in 2020 by Akash Agarwal, Pibit developed an underwriting technology suite called CURE designed to reduce the enormous amount of time insurers spend processing commercial insurance submissions.
For Agarwal the two essential blind spots in post-AI risk are governance and data provenance. “Many organizations are treating AI adoption primarily as a workflow acceleration initiative instead of a risk governance challenge,” Agarwal told Startup Beat. “That means investment goes into models and automation, but not enough into explainability, auditability, model drift monitoring, or understanding where systems fail.”
Jonathan Selby, Technology Practice Lead at Founder Shield, also echoed the need for governance and the need to address fast moving risks. “Effective governance means treating data quality as a security priority and ensuring models are trained on accurate, unbiased information.” Selby told Startup Beat, adding that “[T]o address fast-moving risks, companies need dynamic insurance policies and proactive incident response plans that evolve with their technology.”
AI systems also inherit assumptions and biases according to Agarwal, and these can be embedded in historical underwriting data. “If legacy decisions contained inconsistencies or structural bias, scaling those decisions through AI only compounds the problem. Responsible adoption requires continuous validation, transparent sourcing and strong human oversight.”
The central challenge for companies operating in insurance is evaluating risk accurately and quickly. “Accuracy is very important in the long run,” Head of Marketing Prakar Mohan at Pibit said about the essential balance in determining risk. “There’s a very thin line between being technologically savvy and blinded by technology.”
This becomes especially critical in commercial insurance where policies can involve massive exposures. “Commercial insurance can go very well in millions of dollars,” Mohan said. A trucking fleet with 5,000 vehicles or a factory employing thousands of workers creates enormous potential liabilities. “For each of them, the ticket size is so huge that risk evaluation becomes very important.”
In the post-AI economy, risk is no longer confined to balance sheets or historical claims data. It exists in online behaviour, autonomous systems, hidden reputational signals and rapidly evolving infrastructures that even insurers struggle to fully understand. The future of insurance may ultimately depend less on predicting certainty and more on adapting continuously to uncertainty itself.
