Brokers placing AI-using clients face a market split on coverage with no clear standard yet
Clients deploying artificial intelligence are almost certainly insured for it. The problem is that nobody, not the carrier, not the client, and often not the broker, can say with confidence which policy covers what.
That ambiguity is the central finding of a 2026 RAND Corporation report, The Insurability of Artificial Intelligence, authored by researchers Sasha Romanosky and Celine Robinson. Drawing on public AI incident data, US litigation records, state legislation, and admitted-market insurance filings, the report maps a market split three ways: a minority of carriers affirmatively covering AI losses, a growing number filing broad exclusions, and the majority remaining silent. It is that silent majority which creates the most immediate problem.
Silent coverage is not actual coverage
When a policy says nothing about AI, coverage depends on how a claim is characterized at the time of loss. A hallucinating chatbot that gives a customer incorrect financial advice could be framed as a professional error, a cyber incident, a product defect, or a general liability claim. Different policies respond to each framing differently, and carriers did not write that language with AI in mind. Two businesses running identical AI systems may therefore reach opposite outcomes at claim time. The outcome turns on which carrier wrote the policy and how that carrier interprets the loss.
The RAND report found that AI-related losses span at least 11 insurance lines, among them technology errors and omissions (E&O), professional liability, cyber, directors and officers (D&O), and commercial general liability. A single AI event can simultaneously resemble a cyber incident, a professional error, and a product defect, which means disputes over which policy responds are likely before any claim is settled.
Carriers are already moving to resolve that ambiguity on their own terms. Verisk‘s Insurance Services Office (ISO) made three optional generative AI exclusion endorsements available for commercial general liability policies in January. Those forms, CG 40 47, CG 40 48, and CG 35 08, exclude bodily injury, property damage, and advertising injury connected to generative AI outputs. ISO’s standardized policy language appears in more than 80 percent of US property and casualty policies, according to RAND, so the reach of those endorsements is substantial wherever carriers adopt them.
Berkley has gone further with language across its specialty lines that bars coverage for nearly any claim tied to AI use, development, or deployment, including a company’s own statements and disclosures about how it uses AI. The RAND report also cited reports that AIG, Great American, and Chubb were moving toward similar positions, though adoption varies by carrier and state.
Exclusion surge at renewal
The surge in exclusion activity began in summer 2025, according to RAND’s analysis of admitted-market filings, concentrated in commercial umbrella and commercial general liability policies. The practical consequence at renewal is that a client previously covered implicitly across cyber, tech E&O, and general liability may now find each of those lines quietly narrowed. The hidden AI liability exposure building inside conventional policies is drawing attention across the market, with carriers shifting from passive silence toward explicit affirmative warranties or absolute exclusions.
The RAND report also identified five accumulation mechanisms that could produce correlated losses across many insureds at once. Universal attacks exploit the same vulnerability across many AI systems. Common model dependency arises when many clients share the same underlying infrastructure. AI also acts as a force multiplier for cyberattacks. A legal or regulatory shock can simultaneously render widespread AI practices actionable across many firms. And slow model degradation produces claims across multiple lines and carriers before any pattern is visible, with no single triggering event.
RAND’s recommendations call on state regulators and the National Association of Insurance Commissioners (NAIC) to develop a standardized AI Coverage Notice requiring carriers to declare, for each line, whether AI-related losses are covered, excluded, or left silent. That structure does not yet exist, so the current picture remains carrier-by-carrier and line-by-line. The AI insurance market’s response to D&O exposures shows how unevenly that process is playing out, with insurers monitoring closely but stopping short of sweeping restrictions in most management liability lines.
The underwriting data problem
One reason the market has not coalesced around a standard approach is that carriers have almost no usable claims data. AI losses are only beginning to generate filed claims, legal theories are still forming, and the underlying models change faster than loss patterns can stabilize. That uncertainty is likely to push carriers toward narrower terms, higher retentions, and lower limits as a precaution rather than as a signal that AI risk is well understood. Until a shared data taxonomy exists across carriers and reinsurers, underwriters will be pricing AI exposure largely by feel.
That gap is already visible in the specialty market. Carrier appetite for AI liability coverage ranges from cautious to optimistic, and no AI product has yet found the carrier support needed to scale. A handful of specialty insurers have begun offering affirmative coverage: Munich RE covers both first- and third-party AI losses across hallucinations, bias, privacy violations, and IP claims; the Artificial Intelligence Underwriting Company (AIUC) provides up to $50 million in primary and third-party cover; and newer entrants including Armilla and Testudo target specific deployer exposures. Take-up remains concentrated among technology-sector clients.
The RAND report’s core argument is that affirmative cover, broad exclusions, and carrier silence are not inherently a problem. Different carriers can hold different views on a poorly understood risk. What creates the problem is that neither policyholders nor their brokers can tell the difference at placement time.