Finding 002Open, revised as forms are added

Your AI policy might be excluded too. Two exclusions, two definitions of AI.

Question: What the AI exclusions now on the market actually reach, and whether "AI" means the same thing in each
Short answer: No. Berkley's management liability form reaches any inferring system and a deficient policy; ISO's general liability forms reach generative AI only.
Basis: One specimen form read in full; three ISO forms as described by the Big "I" Virtual University — the ISO form text itself was not read
Tier: Authored here — a comparison of form wording, not a coverage opinion
Scope reviewed: Management liability and CGL — no agents E&O form with an AI exclusion was found
Reviewed: 16 September 2026

The finding

The exclusion most people have heard of is the ISO one. The exclusion most agencies should read is the other one.

Verisk's general liability endorsements (CG 40 47, CG 40 48 and CG 35 08, effective January 2026) exclude injury "arising out of generative artificial intelligence," defined as a machine-based system trained on data with the ability to create content. That is a narrow definition, and it sits in the CGL, which insures the business's own liability to third parties.

Berkley's form PC 51380 00 (06-24) sits somewhere else, in the management liability suite (D&O, employment practices, fiduciary), and it is wider on every axis. Its definition covers any machine-based system that infers from input how to generate predictions, content, recommendations or decisions. Generated content is one example, not the boundary. A rating tool, a triage model, a recommendation engine could each meet that definition, depending on what it does.

Three clauses worth reading twice

(1)(c) any Insured's inadequate or deficient policies, practices, procedures, or training relating to Artificial Intelligence or failure to develop or implement any such policies, practices, procedures, or training

Berkley PC 51380 00 (06-24), Section III exclusions, as amended

The form your E&O carrier tells you to protect yourself with is the exact thing this clause excludes when it falls short. There is no safe harbor for having a policy; only a claim that does not arise from it.

(1)(b) any Insured's actual or alleged failure to identify or detect content or communications created through a third party's use of Artificial Intelligence

Same form

This one reaches something the insured does not control: whether the other side used AI, and whether you noticed. A submission drafted by a client's chatbot and passed through unread is inside it.

"Artificial Intelligence" means any machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments

Same form, definitions (extract)

Whether a particular tool meets that definition is a question about the tool, and the form does not say every ordinary calculation is AI. But it plainly is not limited to things that write text.

Side by side

Two families of exclusion, as read and as described

Form	Coverage part	Edition	Definition of AI	Read in full
Berkley PC 51380 00	Management liability (D&O, EPL, fiduciary), all liability coverage parts	06-24	Any machine-based system that infers outputs from input	Yes, specimen
ISO CG 40 47	CGL Coverage A and B	Jan 2026	Generative AI only	No, as summarized
ISO CG 40 48	CGL Coverage B only	Jan 2026	Generative AI only	No, as summarized
ISO CG 35 08	Products/completed operations, Section I	Jan 2026	Generative AI only	No, as summarized

The same Berkley text was presented to the Lloyd's Market Association Wordings Forum on 21 April 2026 alongside competing definitions from the EU AI Act and others — evidence that no settled market definition exists yet.

What this does not establish

It does not establish that your agency's professional liability policy excludes AI. We searched for an agents E&O form with an AI exclusion or an affirmative AI grant and did not find one; that is a search result, not proof one does not exist. It does not establish that either family is attached to any particular program. A specimen is a specimen.

One affirmative product was announced on 24 June 2026. Mayflower Specialty and Hadron's management-liability AI program, with no form number, edition date, definition or effective date in the release. We will read the form when there is one to read.

What to do with this

Pull the endorsement schedule on every liability policy in your own program and on the programs you place, and read the definition before the operative clause. The definition decides the width. Then ask the carrier one question: does the AI definition on this form reach tools that make recommendations, or only tools that generate content?

Sources: Berkley Insurance Company form PC 51380 00 (06-24), specimen hosted by Hunton; Big "I" Virtual University, "Verisk to roll out new general liability exclusions for generative AI exposures," 21 October 2025; DAC Beachcroft, "AI and Policy Wordings," LMA Wordings Forum, 21 April 2026; BusinessWire release via Morningstar, 24 June 2026. The ISO form text was not read; the descriptions above are the Big "I" summary of the Verisk filing.