Underwriting the Agent Economy: The Blueprint for an AI Insurance Stack

Key Takeaways

• Insurance has been the quiet enabler of major economic and technological developments — from maritime trade to commercial nuclear power — by limiting downside, pricing risk, spreading best practices, and assuring compensation to harmed parties. The AI agent economy, projected to be handling trillions of dollars’ worth of transactions by 2030, is the next such development whose risks insurers must price, manage, and absorb.
• Today frontier AI agent risk is largely unpriced and invisible. The bulk of insurers’ exposure sits as silent coverage inside existing policies (especially cyber, professional, and general liability). Exposure is growing rapidly: enterprise spend on frontier AI grew over 300% in 2025 and looks to accelerate in 2026. Insurers risk repeating the costly ambiguity that plagued cyber insurance, and ignoring a potential time bomb. Nearly half of surveyed Lloyd’s underwriters think their policyholders manage AI risk adequately, while only one in five businesses report a mature governance model for autonomous agents.
• Insurability is trending the wrong way. Agent capability gains are outpacing reliability gains; the upper bound of incident severity appears to be rising, from hallucinated refund policies in 2022 to wrongful death cases in 2025. Furthermore, with over 80% of deployments depending on just three foundation model providers, correlated losses pose a serious risk.
• Conventional playbooks will not work. The length of tasks agents can autonomously complete is doubling roughly every four months: frontier AI is too fast-moving for actuarial models that lag behind reality. This also means actuarial data moats will erode quickly.
• Affirmative AI coverage with limits in the billions is achievable by 2030 — but only if the industry coordinates to build shared infrastructure. This report describes an eight-component AI insurance stack spanning incident data collection, CAT modeling, standards, contract design, risk selection, pricing, monitoring, and claims management. Of note: underwriting can leverage forward-looking performance evaluations, similar to pen-testing in cyber. Many components are complements, posing a cold-start problem; others are public goods: building this stack will require industry-wide coordination.
• The industry has successfully coordinated before. Insurers founded Underwriters Laboratories in 1894 to manage the new hazards of electricity. The Closed Claims Project turned pooled medical malpractice data into best practices that helped drive a sharp reduction in fatalities while expanding insurance availability. AI insurance should borrow from these and other precedents.
• The stakes are large. Handled well, AI insurance will enable responsible adoption and a more resilient economy; handled poorly, the insurance market will stall and a coverage gap will balloon. In the worst case, a catastrophe with only ~$100 billion in direct damages could erase several trillion in GDP if it triggers an economic slowdown, magnified by insurers suddenly withdrawing coverage — just as the collapse of terrorism insurance froze construction and grounded aviation after 9/11.
• The tails require alternative risk transfer. Covering societal-scale frontier AI risks — CBRN, critical-infrastructure collapse, loss of control scenarios — will require purpose-built institutions, potentially including a dedicated mutual for frontier AI model developers, catastrophe bonds, bespoke liability regimes, and government backstops.