Insurers Pursue Generative AI Liability Exclusions

Technical documentation increasingly has value beyond internal model governance. Companies in comparable situations commonly need a traceable record of where AI is used, which vendor or internal model supports a workflow, what human-review controls exist, what data flows into the system, and who owns incident escalation. Those artifacts can support risk assessment, procurement review, legal review, and insurance-renewal discussions without assuming that a particular claim is covered.

UPHelp's guidance has focused on reviewing insuring agreements, exclusions, definitions, endorsements, and carve-backs together rather than treating a policy's AI label as dispositive. Fenwick recommends mapping potential claims to individual coverage lines and considering negotiated endorsements or other structures where gaps emerge. UPHelp similarly advises policyholders to scrutinize broad "arising out of" language and seek narrower definitions or targeted carve-backs during renewal negotiations.

AI risk management is often framed as a model-quality, security, privacy, and compliance problem. Emerging insurance exclusions add a commercial control point: organizations using or supplying AI may need to determine whether a loss involving generated content, automated decisions, or AI-enabled services is covered under any relevant policy.

Claims Journal reports that carriers have shown increased interest in three Insurance Services Office, or ISO, endorsements for commercial general liability coverage. Alana McMullin, a partner at Lathrop GPM, told Claims Journal there has been a "major shift" in insurers' treatment of AI-related risk and that carriers are moving quickly to limit exposure. The publication reports she has seen more carrier filings with state insurance regulators seeking approval to use the endorsements.

A single AI-enabled business process can create several legally distinct loss scenarios. An incorrect generated answer may raise professional-liability questions; a discriminatory automated hiring outcome may raise employment claims; a data exposure involving a model or vendor may implicate cyber coverage; and public statements about AI capabilities can create corporate-governance exposure. The relevant policy and exclusions can differ by scenario.

Hub International identifies examples across those lines, including alleged AI-related hallucinations in professional work, discrimination claims involving AI-powered hiring systems, AI-washing-related securities and regulatory exposure, and cyber risks involving AI systems. Rough Notes reports that AIG, Berkley, Chubb, Great American, and QBE have developed endorsements or provisions restricting coverage for AI claims. These reports do not establish that every insurer uses the same wording or that all AI-related claims are excluded.

Fenwick warns that coverage erosion can occur through narrower definitions and carve-backs, rather than one highly visible exclusion. It characterizes the resulting issue as "gap risk," where no single policy clearly responds to a claim spanning traditional insurance lines.

Broad exclusions can make the distinction between an AI feature and an AI-dependent service commercially significant. Teams building generative AI into customer-facing or high-consequence workflows may therefore benefit from involving risk, legal, security, and procurement stakeholders early, particularly when third-party model providers, agents, or automated decision systems are part of the architecture.