AI risks make some insurers wary of corporate liability

AI-related insurance coverage is becoming more restrictive as insurers confront rising liability risks from misinformation, deepfakes, privacy breaches, IP disputes, fraud, and discriminatory decisions. W. R. Berkley has added AI exclusions to D&O, E&O, and fiduciary-liability products, while Verisk/ISO introduced optional generative-AI exclusion language in January 2026 that could affect forms used across more than 80% of U.S. property-and-casualty policies. RAND cites 713 documented AI incidents and roughly 250 U.S. AI-related lawsuits, arguing that clearer coverage disclosures and a common AI-risk taxonomy are needed as enterprises continue deploying the technology.

Analysis

The investable read-through is less about a near-term liability shock and more about who can monetize uncertainty. Brokers AON, AJG, and BRO should see higher advisory, placement, and renewal activity as enterprises seek bespoke endorsements and negotiate indemnification; they capture the complexity premium without retaining correlated model risk. Underwriters that maintain disciplined exclusions may protect reserve adequacy, but widespread carve-outs also shift liability back to corporate balance sheets and can slow adoption in regulated end-markets.

For enterprise software, the pressure point is not generic AI demand but sales-cycle duration and contract economics. Large customers will increasingly require audit trails, human-in-the-loop controls, cyber coverage confirmation, and vendor indemnities, raising implementation friction and potentially reducing AI-feature attach rates for high-multiple application software before it affects headline subscription revenue. Security, identity, data-governance, and observability spend should be relatively insulated because it becomes a prerequisite to deployment rather than a discretionary productivity project.

Consensus is likely too focused on whether insurers can create a standalone AI product. The larger 6-18 month risk is an accumulation problem: a single model, cloud outage, or widely replicated agentic workflow could generate thousands of economically similar claims, prompting reinsurers to tighten aggregate limits even where primary policies remain nominally available. This thesis is falsified if renewals show no increase in AI-specific exclusions, enterprise AI contract cycles remain stable, and insurers disclose no adverse development or higher reinsurance costs attributable to technology-related liability.

Key Decisions for Investors

• Initiate a 3-6 month long AJG or AON / short KIE pair: brokers should benefit from higher placement complexity and commissions while the insurance ETF retains underwriting-tail exposure. Target 8-12% relative upside; exit if commercial renewal pricing and brokerage organic-growth commentary fail to improve over the next two reporting cycles.
• Use a 6-12 month long PANW or OKTA / short IGV relative-value position to express the shift from AI experimentation toward mandatory security, identity, and governance controls. Risk is that software buyers treat AI controls as bundled cloud functionality; cut if PANW/OKTA billings decelerate while IGV AI-related bookings accelerate.
• Keep WRB on a conditional long watchlist rather than buying solely on exclusion language. Add only if upcoming results demonstrate stable retention, favorable casualty pricing, and no meaningful adverse mix shift; the upside is reserve protection and superior underwriting selection, while the key risk is brokers steering accounts to carriers offering broader coverage.
• Reduce exposure to the most valuation-sensitive enterprise AI beneficiaries where incremental revenue depends on rapid deployment by regulated customers; use CRM, NOW, and ORCL guidance for AI attach rates and deal-cycle commentary as the confirmation signal. A broad software de-rating is not yet warranted without evidence of delayed bookings or increased indemnification costs.