Silent AI and the New AI Exclusions in D&O, E&O, EPL and CGL Policies

Silent AI is the uncertainty that arises when an insurance policy neither covers nor excludes losses caused by artificial intelligence, and insurers are now ending that uncertainty by adding AI exclusions at renewal. ISO's standard generative AI exclusions for general liability took effect in January 2026, and at least one insurer introduced a far broader AI exclusion for D&O, E&O and fiduciary liability in 2025. A company that uses AI should therefore read each renewal for new AI wording and replace any cover that has been removed.

What Is Silent AI?

Most D&O, E&O, EPL and general liability policies in force today were drafted before generative AI was in everyday business use, so they are silent on AI in this sense. A discrimination claim over an AI screening tool, for instance, may fall within an EPL policy's insuring agreement (the clause that grants cover), but nothing in the wording confirms it. An insurer facing a large loss it never priced may contest the claim, leaving the company to fight a coverage dispute alongside the claim itself.

The market resolved the same uncertainty once before, when it was known as silent cyber. On July 4th 2019 Lloyd's issued Market Bulletin Y5258, requiring every policy in its market to state whether it covered cyber losses “by either excluding coverage or by providing affirmative coverage,” beginning with first-party property damage policies that incepted on or after January 1st 2020.^[1] As of this writing, the AI equivalent is being driven by individual insurers' filings rather than a market-wide mandate, so it arrives one policy at a time and is easy to miss. It is best to do for AI what Lloyd's required for cyber and establish, policy by policy, whether AI losses are covered, excluded or left silent.

Why Are Insurers Adding AI Exclusions Now?

Insurers are adding AI exclusions because AI-related claims are growing quickly and most existing policies were never priced for them. According to a Gallagher study reported by Insurance Journal in August 2026, AI-related lawsuits rose 978% from 2021 to 2025 and 137% from 2024 to 2025.^[2] Securities suits are following the same path: by the D&O Diary's count, 24 AI-related securities class actions were filed between January 1st and September 23rd 2026, nearly 14% of all new filings in that period.^[3]

The sums at stake are also large: Cornerstone Research's Disclosure Dollar Loss Index, which measures the drop in a defendant's market capitalization at the end of the class period, stood at $529 billion for filings in the first half of 2026, and AI-related filings accounted for $385 billion, or 73%, of it.^[4]

Insurers have moved to limit this exposure: the Financial Times reported in November 2025, as summarized by Insurance Business, that AIG, Great American and W. R. Berkley had sought regulatory approval to limit their liability for AI-related claims, and Insurance Business later reported that W. R. Berkley, Chubb, Travelers, Berkshire Hathaway and AIG had filed to adopt the ISO generative AI forms or their own equivalents by April 2026.^[5][6]

The market is not moving evenly, however: Marsh's Greg Eskins told Business Insurance in April 2026 that “the market has been restrained in taking any drastic actions with either affirmative or exclusionary language, though both exist,” while Gallagher's John Farley warned that if AI exposures become excluded, “we're going to have to figure out where this exposure should be covered.”^[7] Since one insurer may add an exclusion while another stays silent or offers affirmative cover, no buyer should assume that last year's wording still applies.

What Does ISO Form CG 40 47 Exclude?

ISO form CG 40 47 removes cover under a commercial general liability (CGL) policy for bodily injury, property damage and personal and advertising injury arising from generative AI. It is the broadest of three generative AI exclusion endorsements (standard amendments attached to a policy) that Verisk's Insurance Services Office (ISO) announced in October 2025, and all three took effect in January 2026.^[8][7] ISO defines generative AI, in substance, as a machine-based learning system or model that is trained on data and can create content or responses such as text, images, audio, video or code.^[8]

Coverage B is where a CGL policy normally responds to defamation, privacy and advertising infringement claims, so a company that publishes AI-generated marketing content, or whose chatbot makes statements about other people, can lose that cover once CG 40 47 or CG 40 48 is attached. All three forms address generative AI only, which means a claim arising from a predictive model, such as a pricing algorithm, falls back into silence unless the insurer has written broader wording.

Mayflower Specialty writes D&O, EPL and E&O coverage rather than general liability, so the CGL gap is one to raise with the broker at each renewal: whether one of these forms has been attached and whether the E&O or media liability policy picks up what it removes.

How Are AI Exclusions Appearing in D&O, E&O and EPL Policies?

AI exclusions are reaching D&O, E&O and EPL policies one insurer at a time, usually as new wording added at renewal. They matter more than the general liability forms because these policies are written around the insured's own conduct, so an exclusion keyed to AI use can remove cover for a company's central exposures.

Does My D&O Policy Exclude AI?

A D&O policy may exclude AI if new wording was added at the last renewal, and the first broad example could reach almost any claim with a connection to AI. In 2025 W. R. Berkley introduced what Hunton Andrews Kurth called “the first so-called ‘Absolute’ AI exclusion,” which Insurance Journal reports applies to D&O, E&O and fiduciary liability.^[9][2] It bars payment of loss on any claim “based upon, arising out of, or attributable to: (1) any actual or alleged use, deployment, or development of Artificial Intelligence by any person or entity,” and its later clauses separately name statements and disclosures about AI, violations of laws regulating AI and demands to investigate AI risks. Hunton concluded that “the potential breadth of this exclusion cannot be overstated.”^[9]

Three kinds of D&O claim can be said to arise out of a company's AI or its statements about it: AI-washing securities suits alleging that the company overstated its AI, shareholder derivative suits alleging that the board failed to oversee AI, and investigations into the company's AI statements. An exclusion of this kind can therefore remove cover for the whole category, including defense costs where the policy's definition of loss includes them. Hunton expected the exclusion to be used first against shareholder suits alleging AI-related misrepresentations, but also cautioned that “the devil is in the details, and the exclusion's purported reach is far less certain,” so how far it extends in a given claim is likely to be disputed.^[9] Boards should ask the broker directly whether the D&O renewal contains AI wording, and AI-D&O is the Mayflower module written for this exposure.

Do E&O and Technology E&O Policies Exclude AI?

Some E&O policies now exclude AI, and professional liability is the line underwriters rate as most exposed. In a Lloyd's Market Association survey of 144 respondents in mid-2025, 94% of them underwriters, professional indemnity (the UK term for professional liability) scored highest for impact in AI loss scenarios, followed by cyber.^[10] For a firm whose service increasingly runs through AI, an E&O exclusion for the use of AI can remove cover for its main professional risk, so the renewal should confirm that the policy's professional services include AI-enabled services and that any AI exclusion is limited to defined harms.

Do EPL Policies Exclude AI?

Most EPL policies are still silent on AI, although exclusions have begun to appear there too. Insurance Business reported in July 2026 that, according to Burns & Wilcox broker Lucas Roberts, AI exclusions “appear in closer to 10% of the EPL market,” with the wording generally placed in the common terms and conditions of a management liability package. Travelers' Chris Williams added that “the vast majority of EPL policies do not make a distinction between wrongful employment practices committed by humans versus AI.”^[13]

Both the silence and the new exclusions matter because hiring claims over AI are multiplying. In Mobley v. Workday, an age discrimination case over Workday's algorithmic screening, a federal court in California preliminarily certified a nationwide collective of applicants aged 40 and over on May 16th 2025, and in June 2026 it rejected most of Workday's motion to dismiss the amended claims.^[11][13] In January 2026 applicants also sued Eightfold AI on the theory that AI-generated candidate assessments are “consumer reports” under the Fair Credit Reporting Act (FCRA).^[12]

An EPL renewal should therefore answer three questions: whether a package AI exclusion applies to the EPL part as well as the D&O part, whether the form excludes or sublimits FCRA claims and whether cover extends to a claim arising from a vendor's screening tool.

How Does “Arising Out Of” Wording Widen an AI Exclusion?

“Arising out of” wording widens an AI exclusion because it lets an insurer argue that a claim is excluded even where AI played only a small part in the facts. This opening phrase, the lead-in, sets how close the connection to AI must be, and Hunton noted that the Berkley exclusion's title suggests it is meant to apply “to virtually any claim with a connection to AI.”^[9] An insurer could argue under such wording that an AI-washing suit arose from the company's development of AI, whereas an exclusion limited to claims alleging that AI output caused the loss would be much harder to apply to that suit.

The phrase “by any person or entity” widens the exclusion further by reaching AI used by vendors and customers. How broadly a court would read either phrase is not settled, so both should be the first things a broker is asked to narrow, before a claim rather than after it.

How Should You Read an AI Exclusion?

An AI exclusion should be read for four things, each of which changes the cover the company holds: how it defines AI, how far its lead-in reaches, what it gives back and how it relates to the answers on the application.

1. The definition of AI: A definition limited to generative AI, as in the ISO forms, leaves scoring and screening models silent, while one written around machine learning in general can reach analytics and automation tools the company does not think of as AI. Hunton found the Berkley definition “subject to a myriad of interpretations and perhaps incapable of comprehension for all but the most sophisticated AI engineers.”^[9]
2. The lead-in and whose AI it covers: “Arising out of” and “by any person or entity” are the broadest choices, as explained above, and narrower wording is worth negotiating for.
3. Carve-backs and sublimits: A carve-back restores cover for named claims, such as securities claims against individual directors or defense costs, but one subject to a sublimit (a lower cap inside the overall limit) set far below the policy limit can leave most of the exposure uninsured.
4. The application: Hunton advises policyholders to answer AI questions on applications carefully, because the pace of AI deployment can make even careful answers out of date before the next renewal.^[9] An answer that no longer reflects the company's AI use may give an insurer grounds to contest a claim.

The practical test is to read the exclusion against the company's most likely AI claims and ask the broker, in writing, whether each one would still be covered.

What Are Your Options When an AI Exclusion Appears?

There are three options: negotiating a carve-back or removal with the existing insurer, buying affirmative AI coverage and adding a difference-in-conditions layer over the existing tower (the stack of primary and excess policies). All three depend on evidence of how the company governs AI, such as an inventory of AI systems, a written governance policy and testing records, which the guide to how underwriters assess AI risk explains.

Option 1: Negotiate a carve-back or removal

The first option is to ask the existing insurer to remove the exclusion, narrow its definition or lead-in, or carve back the claims that matter most. It keeps the program in one place and is likely to be the least expensive route, but the result depends on the insurer's appetite and may be partial. Negotiation should be the first step at every renewal, started early enough that the other options remain open if it fails.

Option 2: Buy affirmative AI coverage

Mayflower writes affirmative AI coverage in three modules, AI-D&O, AI-EPL and AI-E&O, on a claims-made-and-reported form (a claim must be both made and reported during the policy period) and on A- (Excellent) AM Best rated paper backed by some of the world's largest reinsurers. The modules are placed through brokers and underwritten on the applicant's AI governance. The benefit is that the policy states in advance which AI exposures it is written for, while the downsides are a separate application, additional premium and coordination with the existing program, including how the retroactive date (the date before which wrongful acts are not covered) lines up with the expiring policies. This option suits a company whose exclusion cannot be negotiated away or whose business depends on AI.

Option 3: Add an AI DIC Excess layer

AI DIC Excess is written to sit over an existing D&O, EPL and E&O tower and to respond, depending on its wording, where the underlying policies are silent on AI or exclude it. It keeps existing insurers and limits in place, but how it attaches and when it responds depend on its own terms, and its cost can only be compared once quotes are in hand. It suits a company whose program is otherwise sound apart from its AI wording.

For most companies the sensible sequence is to negotiate first and then fill what remains, and the coverage gap check shows where a program currently stands.

What Should You Check for AI Exclusions at Renewal?

A renewal review for AI exclusions should cover every policy in the program, because an AI exclusion in one line can push a claim onto another, and it works best when it starts at least 90 days before the renewal date.

The recommended course of action is to ask the broker for a comparison of the expiring and renewal wordings for every line, flagging any new AI definition, exclusion, carve-back or sublimit. Where AI cover has been removed, or left silent on an exposure that matters, it is a good idea to start on the alternatives early: request a quote for the relevant Mayflower modules, or talk to the team about how AI DIC Excess would sit over the current tower.