Insurers must preserve trust
Insurers are being warned that adopting AI depends on preserving customer and regulator trust — not just chasing speed or accuracy. Carrier Management argues that reproducibility, explainability and clear controls are prerequisites for deploying generative or agentic AI in underwriting and operations (carriermanagement.com). Industry data shows AI liability is emerging like cyber risk for SMEs — 74% of small businesses already use AI and 91% expect to adopt it soon — which raises new exposure insurers must model and price (insurancebusinessmag.com). Vendors and regulators are responding: there's active vendor guidance on building GRC automation and a legal webinar series on generative-AI coverage questions, signalling the issue is moving fast toward compliance and litigation scrutiny (drata.com, natlawreview.com).
Insurance companies are getting told to slow down on one part of artificial intelligence: not the software itself, but the trust around it. A Carrier Management piece published April 9 says insurers can’t treat generative artificial intelligence like a faster calculator if they can’t show regulators and customers how a decision was produced. (carriermanagement.com) That warning lands as insurers push artificial intelligence into underwriting, claims, servicing, and sales. Carrier Management says the pressure is no longer just about speed or accuracy, because privacy, cybersecurity, and explainability now sit in the middle of every deployment. (carriermanagement.com) In insurance, underwriting is the step where a carrier decides whether to take a risk and what price to charge for it. If an artificial intelligence system helps make that call, the insurer needs to reproduce the path it took the way an accountant can reproduce a spreadsheet. (carriermanagement.com) That is why reproducibility keeps coming up. A model that gives one answer on Monday and a slightly different answer on Tuesday for the same file creates a regulatory problem before it creates a technology problem. (carriermanagement.com) Explainability is the second pressure point. If a customer is denied coverage or quoted a higher premium, state examiners and internal compliance teams will want more than “the model said so” as a reason. (carriermanagement.com) The backdrop is that small businesses are already using artificial intelligence at cyber-risk scale. Insurance Business reported April 10 that 74 percent of small and medium-sized businesses already use artificial intelligence, and 91 percent say they expect to adopt it soon. (insurancebusinessmag.com) That is starting to create a new insurance category. The same report says brokers and underwriters are treating artificial intelligence liability the way the market once treated early cyber risk: a fast-growing exposure with unclear policy language, uneven controls, and losses that may surface before pricing catches up. (insurancebusinessmag.com) Insurers are already moving from theory to product. On March 18, specialty insurer Hartford Steam Boiler, part of Munich Re, announced artificial intelligence liability coverage for small businesses, including claims tied to bodily injury, property damage, and advertising injury linked to artificial intelligence use. (munichre.com) Vendors are moving in parallel by selling governance tools instead of just model performance. Drata’s insurance-focused guidance pitches governance, risk, and compliance automation as a way to cut audit preparation, reduce manual risk work, and keep compliance continuous instead of point-in-time. (drata.com) Law firms are moving too, which usually means a market expects disputes. Wilson Elser’s May 2026 Insurance Coverage Master Class includes a session on generative artificial intelligence and future coverage litigation, a sign that policy wording and claim fights are already becoming a billable specialty. (digital.wilsonelser.com) Regulators are closing in from the other side. Plante Moran wrote on March 24 that the National Association of Insurance Commissioners is considering updates to its artificial intelligence model bulletin and that insurers should expect governance, cybersecurity controls, and documentation to stay a regulatory priority. (plantemoran.com) So the fight inside insurance is no longer “use artificial intelligence or don’t.” It is whether carriers can make artificial intelligence behave like regulated infrastructure, with logs, controls, geography limits, testing, and records sturdy enough to survive an examiner, a plaintiff lawyer, and a customer complaint. (carriermanagement.com)
