W.R. Berkley has attached an absolute AI exclusion to its directors and officers, errors and omissions, and fiduciary liability forms, a scope none of the ISO general liability endorsements reach. S&P Global Market Intelligence counted 41 P&C groups that have filed at least one subsidiary to exclude generative AI, with 20 more filing to delay (The Insurer, July 2026).

The question is not whether carriers keep excluding AI. It is whether a management liability book can price a tail it has declared it will not model.

Key Takeaways

  • PC 51380 treats inadequate AI governance as an independent trigger. A board sued over an AI strategy it approved without documented oversight can lose D&O coverage even if no model produced a bad output.
  • AI cases were 13% of core securities filings in the first half of 2026 but carried $385 billion of the $529 billion disclosure dollar loss index, 73% of the total.
  • Securities class action filings rose 30% to 121 in the half, with 15 of them AI-related.
  • The market is mid-decision, not settled: 41 groups have filed to exclude and 20 have filed specifically to delay, keeping AI claims inside their books another renewal cycle.
  • Public-company D&O rates rose 2% to 5% in 2025 in an otherwise soft market, so the exclusion exits the one pocket where rate is moving in carriers' favor.

Two Exclusions, Two Different Perils

The wave traces to one filing. Verisk's Core Lines Services made a multistate submission in July 2025 including an optional generative AI exclusion for commercial general liability, and CG 40 47, CG 40 48 and CG 35 08 began attaching to CGL renewals effective January 1, 2026. CG 40 47 strips bodily injury and personal and advertising injury for any loss arising out of generative AI; CG 40 48 removes only the advertising injury piece; CG 35 08 reaches products and completed operations. All three use "arising out of," which needs only a causal link, and they cleared roughly 80% of state reviews.

Berkley's "Artificial Intelligence Absolute Exclusion," Form PC 51380, bars coverage for any claim "based upon, arising out of, or attributable to" the actual or alleged use, deployment, or development of artificial intelligence, by any person or entity, across D&O, E&O and fiduciary liability (National Law Review, May 2026). Bloomberg Law called it insurance's first-of-its-kind absolute exclusion reaching board-level coverage.

FormLines coveredWhat triggers the exclusion
CG 40 47 / CG 40 48 / CG 35 08 (Verisk/ISO)CGL, products/completed operationsBodily injury, property damage, or advertising injury arising out of generative AI use
PC 51380 (W.R. Berkley)D&O, E&O, fiduciary liabilityAny claim arising out of AI use, deployment, development, governance failure, or non-disclosure, regardless of whether AI caused the underlying harm

The scope difference is the substance. The CGL forms exclude harm caused by AI: a defective product, a defamatory chatbot, a factory injury. PC 51380 excludes a board's own decisions about AI, reaching shareholder derivative suits over inadequate governance, securities claims alleging misrepresented AI capability, fiduciary claims tied to AI-driven plan administration, and a company's failure to detect or disclose third-party AI use in its own filings.

That governance clause does not require a malfunction. "Given that everything including our phones with Google now uses AI, I don't know how anything would be left if there is a broad AI exclusion," said Reed Smith partner Courtney Horrigan, for policyholders (Bloomberg Law, July 2026).

Why a D&O Triangle Does Not Look Like a GL Triangle

General liability AI claims are idiosyncratic: one chatbot, one output, one injured party, reported and often resolved within a few years of the policy period. That is what the CGL endorsements were built to carve out.

Management liability claims behave differently in both dimensions that matter to a reserve. They are long-tail, commonly three to seven years from the underlying event to resolution once discovery, class certification and appeals run. They are correlated rather than idiosyncratic: one disclosure about a company's AI capabilities generates a single claim naming dozens of directors and officers, with damages scaling to the whole shareholder class.

Cornerstone Research's midyear count sizes both. Securities class action filings rose 30% to 121 in the first half of 2026, and 15 of those, 13% of core filings, were AI-related. That 13% carried $385 billion of the $529 billion disclosure dollar loss index for the half, 73% of the total (Cornerstone Research, July 2026). "AI-related cases represented a modest share of total filings but an outsized share of alleged investor losses," said Stanford's Joseph Grundfest, a former SEC commissioner.

Frequency-modest and severity-dominant breaks a loss-cost model built on idiosyncratic GL events. It is also the profile a D&O actuary has reserved before, through the tech-bubble and financial-crisis filing waves, which is the argument for rating the exposure rather than removing it. Verisk liability vice president Joe Lam framed the GL logic for standardized forms as preserving the alternative: "Without exclusions to allow underwriters a level of stability or to accept a risk, you run into a situation where they might just walk away from the risk."

An absolute exclusion says the tail cannot be bounded well enough to hold capital against it at any market price. Pricing it says the tail is wide but bounded. The market is currently split on which: 41 groups filed to exclude, 20 filed to delay, and public-company D&O rates rose 2% to 5% in 2025 while private D&O held flat to down 5% on more than half of renewals (Founder Shield, January 2026). The one pocket of hardening in a soft market is the segment PC 51380 exits.

The Exclusion Does Not Reach the Policies Already Written

Whatever a carrier decides for the next renewal, the reserving exposure sits on business already earned.

D&O and fiduciary policies written in 2024 and 2025 carry no AI exclusion and remain exposed to AI-related claims that have not been reported, given the multi-year gap between an alleged governance failure and a filed securities suit. Silent AI exposure in a general liability book is a known problem; the management liability version is larger in dollar terms and slower to surface, and the exclusion cannot bound it because it was not in force when those policies were written.

The exclusion's own language adds a second layer for business written after it attaches. Broad, novel exclusionary wording invites coverage litigation over what "arising out of" AI means for a specific claim. A director sued over a decision informed by, but not solely caused by, an AI-assisted analysis is the ambiguous fact pattern the phrase was drafted to sweep in, and the same pattern a court reading "arising out of" broadly, as courts have for decades in general liability disputes, may also sweep in. Where a court narrows it instead, coverage attaches on a claim the pricing never treated as an AI risk.

Either outcome lands in the same place on a reserve review. The tail sits inside loss development factors selected before an AI-correlated claim class existed, on accident years whose policies contain no exclusion to cap it. Cornerstone's severity data describes a class that is already disproportionate, not one that might become so.

Further Reading

Sources