Summary Coverage language is where the artificial intelligence governance debate stops being theoretical and starts being priced. Carriers are showing growing interest in three new ISO exclusions aimed at generative AI risks in commercial general liability policies, set against AI-related litigation that rose 978% between 2021 and 2025.
Source: Claims Journal · 20 July 2026 · read the original article
What is being excluded, and why The three exclusion forms respond to the claim types that dominate the emerging AI docket: patent infringement, copyright infringement and privacy violations. Each sits awkwardly inside a traditional general liability wording. Personal and advertising injury cover was drafted for defamation and trade dress disputes, not for training-data provenance. Bodily injury and property damage triggers were drafted for physical loss, not for an automated decision that misprices a customer or denies a claim.
The result is a coverage grey zone that insurers can resolve in one of two directions: exclude the exposure and leave it uninsured, or affirmatively underwrite it with defined conditions. The market is currently split, and that split is itself informative — it signals that carriers do not yet share a view on whether AI loss is quantifiable enough to price.
The underwriting question behind the exclusion An exclusion is a decision made under uncertainty about frequency and severity. Underwriters facing a novel exposure with a thin loss history have three levers: exclude, sub-limit, or condition cover on demonstrable controls. The third lever is the one that matters for risk functions, because it converts governance maturity into premium.
The controls that are likely to become underwriting questions are the same ones supervisors are asking about: a maintained inventory of deployed models, documented human review, bias and drift testing on customer-affecting systems, and contractual rights against model vendors. A firm that can evidence these is a different risk from one that cannot, and pricing will eventually reflect that.
What policyholders should do before renewal - Map every AI system that touches customers, pricing, claims or credit decisions, and identify which policy would respond to a resulting claim - Read the current wording for silent AI exposure — cover that is neither granted nor excluded is the position most likely to be litigated - Prepare an AI control narrative for underwriters covering inventory, human oversight, testing and vendor rights - Track intellectual property exposure specifically, since training-data and output-ownership disputes are the leading claim category - Escalate uninsured residual exposure to the risk committee rather than leaving it inside an insurance renewal file
Methodology and limitations This analysis summarises a publicly reported market development as at the date shown and links to the original source. Exclusion forms vary by carrier and jurisdiction, filings continue to evolve, and nothing here is legal or coverage advice. Litigation growth figures are as reported and reflect filed claims rather than adjudicated liability.
Related reading See [Insurance Risk](/expertise/insurance-risk), [Model Risk](/expertise/model-risk), [Enterprise Risk](/expertise/enterprise-risk) and the overview in [AI governance in insurance and banking](/insights/ai-governance-insurance-banking-2026).
Frequently asked questions
What is being excluded, and why?
The three exclusion forms respond to the claim types that dominate the emerging AI docket: patent infringement, copyright infringement and privacy violations. Each sits awkwardly inside a traditional general liability wording. Personal and advertising injury cover was drafted for defamation and trade dress disputes, not for training-data provenance. Bodily injury and property damage triggers were drafted for physical loss, not for an automated decision that misprices a customer or denies a cl...
What should risk leaders know about the underwriting question behind the exclusion?
An exclusion is a decision made under uncertainty about frequency and severity. Underwriters facing a novel exposure with a thin loss history have three levers: exclude, sub-limit, or condition cover on demonstrable controls. The third lever is the one that matters for risk functions, because it converts governance maturity into premium.
What should risk leaders know about methodology and limitations?
This analysis summarises a publicly reported market development as at the date shown and links to the original source. Exclusion forms vary by carrier and jurisdiction, filings continue to evolve, and nothing here is legal or coverage advice. Litigation growth figures are as reported and reflect filed claims rather than adjudicated liability.
What should risk leaders know about related reading?
See [Insurance Risk](/expertise/insurance-risk), [Model Risk](/expertise/model-risk), [Enterprise Risk](/expertise/enterprise-risk) and the overview in [AI governance in insurance and banking](/insights/ai-governance-insurance-banking-2026).