Does your E&O or general liability cover AI? In 2026, probably not anymore.
By Joel Singh · Reviewed by iSL Advisory · Last reviewed 2026-07-04
If your business uses AI in any client-facing way, the honest answer to "am I covered if it goes wrong?" changed on January 1, 2026. It did not change because your business changed. It changed because the insurance industry rewrote the fine print.
This page explains what standard policies used to cover, what the new exclusions remove, who is actually exposed, and a practical framework for deciding what to do about it. This is general information, not insurance advice, and it is not a recommendation of any specific policy. Talk to a licensed broker about your own program.
What E&O and GL covered (before 2026)
- If your agency delivered a flawed report, it usually did not matter to the E&O carrier whether a junior analyst or a language model produced the flaw.
- If your marketing content triggered an advertising injury claim, the CGL policy did not ask what software wrote the copy.
- Coverage existed by omission: the policies were silent on the subject, and silence generally worked in the policyholder's favor.
What the 2026 exclusions remove
- CG 40 47 Bodily injury, property damage, and advertising injury arising from generative AI on CGL policies.
- CG 35 08 Generative AI exposures under products and completed operations coverage.
- WR Berkley "Any actual or alleged use, deployment, or development of Artificial Intelligence" on D&O, E&O, and fiduciary lines.
What your policies quietly covered before 2026
Until recently, most businesses never asked whether their errors and omissions (E&O) or commercial general liability (CGL) policies covered AI, because they did not need to. The policies were silent on the subject, and silence generally worked in the policyholder's favor. If your agency delivered a flawed report, it usually did not matter to the E&O carrier whether a junior analyst or a language model produced the flaw. If your marketing content triggered an advertising injury claim, the CGL policy did not ask what software wrote the copy.
Insurance professionals call this "silent AI" exposure: coverage that exists by omission rather than by design, in the same way "silent cyber" existed before carriers began writing explicit cyber exclusions [src]. Carriers dislike silent exposure because they cannot price it. So they are now doing what they did with cyber: excluding it explicitly, and selling the coverage back as a separate, underwritten product.
The exclusion wave, in plain English
Three things happened in quick succession.
First, the standard forms changed. Verisk (ISO), the organization that drafts the standard policy language most US carriers build on, rolled out new general liability endorsements effective January 1, 2026. CG 40 47 excludes bodily injury, property damage, and advertising injury arising from generative AI on CGL policies. CG 35 08 applies the same exclusion to products and completed operations coverage [src] [src]. These are optional endorsements, not automatic changes to every policy. But they give every carrier in the country ready-made, regulator-reviewed language to attach at renewal, and that is the point. The tools to exclude AI are now on the shelf.
Second, carriers started filing their own versions. At least six major carriers have filed carrier-specific AI exclusions, and more than 80 percent of those filings have been approved by state regulators, with Florida, Connecticut, and Maryland approving fastest [src]. Chubb has been described as leading exclusion filings among legacy carriers [src].
Third, some of the language is absolute. WR Berkley has filed an absolute AI exclusion on directors and officers, E&O, and fiduciary lines that removes coverage for "any actual or alleged use, deployment, or development of Artificial Intelligence" [src] [src]. Read that wording again. Not "AI you built." Not "AI that caused the loss." Any actual or alleged use. Under language like that, a routine claim can be contested simply because AI was somewhere in the workflow.
The practical consequence is coverage fragmentation: two businesses with identical operations can have completely different AI coverage depending on which carrier's paper they renewed onto, and neither may know it until a claim is denied [src].
Who is actually exposed
You do not need to be an AI company to be caught by this. The exposure follows AI usage, not AI branding.
Agencies and professional services firms using AI in deliverables.
If AI contributes to reports, analysis, creative work, code, or advice you deliver to clients, an E&O claim arising from an AI-produced error may now land on excluded ground. Marketing is the sharpest edge here: a survey by HSB found that 74 percent of small and mid-sized businesses already use AI and 91 percent plan to, with marketing as the top use case at 47 percent [src]. AI-generated advertising and marketing content maps directly onto the advertising injury coverage that CG 40 47 now excludes [src].
Software companies with AI features.
If your SaaS product added an AI feature, your tech E&O may or may not respond to claims arising from that feature's output, depending on your carrier's current wording. Standalone products in this market exist specifically because incidents like hallucinations, model drift, and deviations from expected behavior can produce third-party claims where a CGL policy may not respond [src].
Any business with a customer-facing chatbot.
A chatbot that gives a customer wrong information is an AI output causing a third-party loss. Whether your existing liability program responds now depends on which exclusions your renewal carried.
Boards and executives.
The WR Berkley exclusion reaches D&O and fiduciary lines [src], which means AI-related management decisions can also fall outside coverage, not just AI-related operational errors.
A decision framework
Here is the sequence I recommend working through, in order.
Check your renewal documents for the exclusion forms.
Pull your current CGL and E&O policies and any renewal quotes, and search for form numbers CG 40 47 and CG 35 08, plus any endorsement with "artificial intelligence" or "generative AI" in the title. Ask your broker directly, in writing: has this carrier filed or attached any AI exclusion, and does my renewal include one? Renewals through 2026 and 2027 are where these exclusions will actually land on policyholders [src].
Build an AI usage inventory before you shop.
You cannot assess a coverage gap without knowing what AI you actually use: which systems, in which workflows, touching which clients and decisions. This inventory is also the first document underwriters will ask for if you pursue affirmative coverage, so it is not wasted work.
Match the gap to the fix.
Once you know what is excluded and what you use, there are broadly three paths: negotiate the exclusion at renewal (sometimes possible while the market remains competitive), add an AI endorsement to an existing cyber or tech E&O policy, or buy a standalone AI liability product.
Endorsement or standalone policy?
An endorsement is usually the right starting point when AI is a feature of your business, not the business. Several are already in market. Coalition offers an affirmative AI endorsement on cyber policies that expands the definition of a security failure to include AI security events, though its scope is AI as an attack vector rather than liability for AI output [src]. Embroker automatically includes an AI Coverage Endorsement on eligible tech E&O and cyber quotes in its startup program [src]. Vouch distributes an AI insurance offering for startups covering AI errors and omissions, algorithmic bias and discrimination, regulatory investigation defense costs, and IP infringement claims from AI systems [src]. Read scope carefully: an endorsement that covers AI as an attack surface is not the same as one that covers your AI's output harming a third party.
A standalone policy makes sense when AI output is central to what you sell, when contracts demand it, or when your existing program now carries a broad exclusion. The standalone market is real but small: only a handful of true standalone AI liability products exist worldwide as of mid-2026 [src]. Armilla, a Lloyd's coverholder underwritten with Chaucer, launched the first affirmative AI liability policy in April 2025, covering hallucinations, model drift, and critical errors leading to damages [src] [src]. Testudo, launched January 2026 with Lloyd's backing, writes generative AI liability for mid-market US vendors and deployers [src] [src]. Munich Re's aiSure, distributed in partnership with Mosaic since February 2026, pays on breach of a predefined performance threshold rather than a negligence claim [src] [src]. For smaller businesses, HSB announced a standalone AI liability product for SMBs in March 2026, delivered through partner carriers' business policies pending state approval rather than sold direct [src] [src]. And for businesses whose existing program already carries exclusions, Relm's PONTAAI product is built specifically as an excess difference-in-conditions wrap over programs that exclude AI [src].
Most of the standalone products are surplus lines, bought through a specialist broker rather than off the shelf [src]. Expect underwriting to be more involved than a standard renewal.
The part nobody tells you: underwriters now price your governance
Here is the pattern across every product above. Armilla requires an AI system assessment before it will write a policy [src]. AIUC requires certification against its AIUC-1 standard, built on more than 5,000 adversarial tests, before insuring an AI agent [src]. The Geneva Association reports that insurers are tightening underwriting standards by scrutinizing insureds' AI systems and governance practices before granting coverage [src]. WTW's 2026 marketplace analysis points the same direction: underwriting is moving toward AI governance posture, meaning model inventories, documented human oversight, bias mitigation, and failure mode documentation [src].
In other words, whether you can get AI coverage, and what you pay for it, increasingly depends on the governance evidence you can hand an underwriter. Organizations that can produce that documentation get coverage on workable terms; those that cannot face exclusions, sublimits, or premium increases [src].
That documentation set is knowable and finite. We have laid out the nine documents underwriters actually ask for, what each one is, and what good looks like, in our coverage-readiness checklist. If you are heading into a 2026 or 2027 renewal with AI anywhere in your operation, start there before you start shopping.
This page is general information about the insurance market, not insurance, legal, or financial advice. Coverage terms vary by carrier, state, and policy. Consult a licensed insurance broker about your specific situation. Last reviewed: 2026-07-04.