Major insurance carriers are pulling back from covering AI-related harms as incidents multiply and legal exposure grows uncertain, according to a report published this week by the RAND Corporation, a nonprofit research organization. The think tank warns that businesses face growing financial vulnerability from deploying AI systems even as they race to adopt the technology. Companies now confront a mismatch between rapid enterprise AI adoption and a splintered market for insurance protection, particularly in the United States.
The scale of documented AI failures is substantial. An Artificial Intelligence Incident Database tracked 713 incidents drawn from over 6,000 reports at the time the study was conducted, covering uses well beyond chatbots. Misinformation and manipulation led with 586 events, followed by deepfakes and synthetic media at 346, deepfake-enabled misinformation at 333, and hallucination and factual errors at 215. Harmful content generated 92 incidents, while agentic and autonomous failures accounted for 84. Privacy and data leaks totaled 58, bias and discrimination reached 47, copyright and intellectual property issues numbered 20, and wrongful AI attribution stood at seven, with 14 events classified as other or unclassified. Some incidents spanned multiple categories, causing the total to exceed 713. Alongside these documented harms, roughly 250 US lawsuits related to AI have emerged, primarily centered on copyright and intellectual property but also addressing privacy and surveillance, fraud and deception, negligence and product liability, discrimination and civil rights, and contracts or trade secrets.
The report observes that certain insurance companies are now excluding AI-related harms from coverage, with insurer W. R. Berkley having introduced exclusions in its directors and officers, errors and omissions, and fiduciary liability products to bar coverage for "any actual or alleged use, deployment, or development of Artificial Intelligence." In January 2026, Verisk/ISO—whose standardized forms appear in more than 80 percent of US property and casualty policies—introduced optional language carriers can adopt to exclude bodily injury, property damage, and other harms arising from generative AI. During W. R. Berkley's fourth-quarter 2025 earnings call, CEO W. Robert Berkley emphasized underwriters need to grasp "the impact that [new technologies like AI are] having on our insureds, what it means for risk, and our ability to fully understand that risk so we can control it, select it, and price for it."
The insurance pullback stems from what the report characterizes as AI-related harms that don't fit neatly within existing insurance lines—incorrect or misleading outputs, deepfakes, privacy violations, intellectual property disputes, fraud, product defects, and discriminatory decisions. These create demand for coverage but leave insurers uncertain about risk calculations and pricing models. While some carriers are retreating, coverage gaps are being filled by new and existing companies that believe they can handle the risk calculations, the study notes. Businesses continue deploying AI despite the uncertainty over whether their corporate insurance policies actually protect them, complicating their fiduciary obligations.
RAND argues that policy researchers, brokers, carriers, and reinsurers need to develop a common taxonomy to track AI incidents and claims, and the organization wants state regulators to push for an AI Coverage Notice so everyone understands what's covered and what isn't. The think tank expects AI insurance coverage will move beyond being a specialty product once the risks are understood and priced accordingly, but that will take time. If AI ends up being uninsurable, the AI industry will have to moderate its ambitions and sales targets while corporate customers delay AI projects to fulfill their fiduciary obligations. The refusal of traditional carriers to underwrite novel technologies may impose discipline on deployment timelines that regulatory frameworks have so far failed to achieve, forcing enterprises to weigh enthusiasm against balance-sheet exposure in ways that slow adoption to a more measured pace.