SOA Survey: AI Adverse Outcomes Top Insurers' Long-Term Risk List
The Society of Actuaries Research Institute and the Casualty Actuarial Society released the 19th Annual Survey of Emerging Risks on March 10, 2026, from fieldwork run that January. Artificial intelligence adverse outcomes was one of the two most impactful risks the C-suite group named for three or more years out, alongside greater-than-normal financial volatility. The near term looks different: 60 percent of that group put economic and geopolitical risks first for 2026, led by financial volatility and geoeconomic and globalization shifts. The survey drew more than 350 responses, over 100 of them from chief risk officers, chief actuaries, lead partners, and senior thought leaders.
The result is not surprising, but it is worth paying attention to because actuaries are trained to measure risk rather than chase trends. When they rank AI adverse outcomes at the top of that longer horizon, it signals that AI is no longer viewed primarily as a cost-saving or innovation opportunity. It is a risk that can affect balance sheets, reserves, and reputations. That shift in framing is what makes the survey more than a headline.
One split inside the C-suite group cuts against reading that as an industry-wide verdict. Respondents at property and casualty carriers named discrete extreme weather events as most impactful in both horizons, and it was the consulting firms that put AI adverse outcomes first for 2026. Seventy-one percent of the C-suite group gave P&C as their primary practice area.
The survey defines the risk broadly, as loss from the intended or unintended negative consequences of advances in AI on individuals, businesses, ecosystems, or economies. Insurers have been quick to adopt predictive models for underwriting, pricing, claims, and fraud detection. Read against a definition that wide, the part an insurer actually controls is narrower: model errors, data drift, biased outputs, and the difficulty of explaining automated decisions to policyholders and regulators. These are not theoretical problems; they map directly to market conduct exams, litigation, and regulatory enforcement actions that are already happening.
Carriers can read the survey as a prompt to align AI governance with enterprise risk management. Models that affect underwriting or reserving should be treated like catastrophe models or economic scenario generators: subject to validation, independent review, and documented assumptions. The difference is that AI models change faster and often depend on vendor platforms that the carrier does not fully control.
The risk conversation also has to include the people who use the models, not just the data scientists who build them. A claims adjuster who follows an AI recommendation without understanding its limits is a different risk than a model with a coding error. Governance must address both the technology and the workflow around it. Model validation can be clean in isolation and still miss the operational risk that lives in how frontline employees read and act on the output.
If actuaries, who are central to how insurers set reserves and price risk, view AI adverse outcomes as a top risk on the long horizon, then boards and risk committees should expect more detailed questions about AI risk appetite, model risk tolerance, and contingency plans for AI failures. The answer cannot be a generic policy; it needs to be specific about which systems are high-risk, how often they are monitored, and what happens when monitoring reveals a problem.
Every survey now reports that AI carries risk. This one is worth reading for who is saying it: the people who set reserves have started pricing AI as a balance-sheet exposure rather than an IT project, and reserving opinions are where that reclassification surfaces first. The survey does not report what number any of them attached to it, and until AI governance turns up inside a reserving opinion, nobody outside that process will see one.