The AI Liability Insurance Gap: Why Your Policy Probably Doesn't Cover a Generated Ad

What the AI Liability Insurance Gap Actually Is

The AI liability insurance gap is the quiet space between the ad you just shipped and the policy you thought covered it. A commercial general liability policy is supposed to respond when an advertisement causes harm: defamation, copyright infringement, invasion of privacy. But a wave of 2026 policy language now excludes losses that arise out of generative AI, and most teams have no idea the wording changed at their last renewal. The result is a coverage hole that opens exactly where generative video lives.

Consider the scenario nobody budgets for. A brand ships an AI-generated spot. A generated spokesperson says something defamatory, or a generated asset quietly lifts a competitor's copyrighted frame, or a generated product shot makes a claim the company cannot substantiate. Any one of those is a classic advertising-injury claim. Under a pre-2026 policy it would have triggered coverage. Under a 2026 policy endorsed with the new exclusions, it may not. The gap is not theoretical, and it is growing.

For years the question was silent. Older policies neither named generative AI nor excluded it, so a court might have found coverage under existing advertising-injury language. That ambiguity, the industry's 'silent AI' moment, is exactly what the 2026 endorsements resolve: they turn uncertainty into an explicit carve-out. The policyholder who never read the form is now unambiguously outside coverage.

The ISO Endorsements That Changed the Baseline in January 2026

The shift started with three optional endorsement forms from ISO, the standards body behind most US commercial policy language. CG 40 47, CG 40 48 and CG 35 08 became available to carriers on 1 January 2026, and they let an insurer carve generative-AI losses out of a commercial general liability policy. CG 40 47 is the broadest: it excludes bodily injury, property damage and personal and advertising injury arising out of generative AI under both Coverage A and Coverage B. CG 40 48 is narrower, stripping only Coverage B, and CG 35 08 reaches products and completed operations.

The phrase that does the damage is 'arising out of'. In coverage law that test is broad: it needs only a causal connection, not proximate cause. A claim does not have to be caused entirely by AI to fall inside the exclusion; it only has to arise out of the use of a generative tool. For a marketing team, that means a generated ad sitting anywhere in the story of a claim can be enough to push the whole thing outside the policy. Carriers did not invent this out of caution alone; reinsurers pushed primary insurers to resolve the ambiguity before treaty renewal, and the ISO forms standardise that push at the policy level.

Where Generated Ads Actually Fall Through the Gap

Personal and advertising injury is the section where AI content claims actually live. Defamation, copyright infringement and invasion of privacy are all Coverage B exposures, and a generated ad touches all three: a synthetic spokesperson can defame, a generated asset can infringe, a generated script can invade privacy. CG 40 48 was written to strip exactly this section, and CG 40 47 strips it along with everything else. So the most likely AI claim a video team faces is also the one most likely to be excluded.

The case that should end the 'it was just the AI' defense is Moffatt v. Air Canada. In 2024 the British Columbia Civil Resolution Tribunal held the airline liable for negligent misrepresentation after its chatbot gave a customer wrong information, confirming that a company owns what its AI tools tell people. The tribunal rejected the argument that the chatbot was a separate legal entity. For commercial video, the lesson is direct: if a generated ad says something untrue, the brand answers for it as if the words were printed on a billboard.

The same discipline applies to whether the work is commercially rights-safe, because a generated asset that lifts a competitor's copyrighted material is exactly the advertising-injury claim CG 40 48 was written to strip out.

Remember that a generated product shot is a claim: a frame showing the product perform is a promise somebody has to prove, and a denied CGL claim will not cover a promise the policy now excludes.

Why 'It Won't Happen to Us' Is a Weak Defense

The adoption math makes the gap unavoidable. A 2026 survey by HSB, part of Munich Re, found that 74 percent of small and midsize businesses already use AI and 91 percent plan to, with 47 percent using it specifically for marketing. That is the population now attaching AI exclusions to their renewals. The tools moved into the workflow faster than the policy wording did, and the mismatch will not show up until the first denied claim reaches litigation.

Moffatt is the warning here too. Air Canada did not set out to mislead; its chatbot simply produced a wrong answer, and the company was held responsible anyway. A video team that ships a generated ad with a fabricated statistic or a borrowed asset is in the same posture. The exclusion does not care that the error was unintentional, and the court will not let the brand point at the model.

The discomfort for finance is real. A denied claim is not just a lost reimbursement; it is an unexpected legal bill landing on a campaign that was meant to save money. The whole pitch for generative video is lower production cost, and the exclusion quietly moves the risk from a priced insurance line back onto the balance sheet.

The Standalone Market Trying to Fill the Hole

Insurers noticed the hole and started building products to fill it. On 18 March 2026, HSB introduced an AI Liability Insurance that pays for AI-related losses some general liability policies exclude, including bodily injury, property damage and advertising injury from AI-generated advertising, marketing, blogs and social media. It is distributed through carrier partners rather than sold directly, and it is designed to sit alongside a standard policy as affirmative coverage.

HSB is not alone. A small group of specialist carriers and managing general agents now writes standalone AI liability, some with limits reaching tens of millions of dollars, much of it in the surplus-lines market outside standard admitted channels. The capacity is real but young, and most products ask for governance documentation a smaller team may not have prepared. The point for a commercial video team is not to buy a specific policy today, but to know the hole exists and that affirmative coverage is now a purchasable thing.

What an affirmative policy actually wants is evidence of control. Carriers writing AI liability ask for the same things a good production workflow already produces: a record of which models generated which assets, a disclosure and substantiation step before publish, and a rights-clearance packet for every face and frame. Teams that already run a governance playbook walk into that conversation with the paperwork in hand; teams that ship on instinct do not.

A Risk Checklist Before You Ship a Generated Ad

Insurance is the backstop. The front line is the brief. Before a generated ad goes live, run a short coverage-aware check so a denied claim never forms in the first place.

First, read the schedule of forms at your next renewal and ask the broker, in writing, whether any AI exclusion is attached. Second, document where generative AI is used in the production pipeline, because a documented governance posture is what an affirmative policy will ask to see. Third, treat disclosure and substantiation as part of the coverage posture, not just compliance. Fourth, confirm the rights on every generated asset before it ships. Fifth, decide now who owns the claim if a generated frame causes one, and whether that answer survives contact with the policy language.

Building the AI video disclosure rules into the brief is the cheapest control you have, because a labelled and substantiated ad is far harder for a claimant to attack than one that hoped to stay quiet.

The teams that sleep at night are the ones who treat a governance playbook as the front line, documenting where AI is allowed and gating every cut before it ships.

Treat Coverage as Part of the Brief, Not the Fine Print

The AI liability insurance gap is not a lawyer's problem to solve after the fact. It is a production decision that should sit in the brief alongside the creative ask, because the ad that triggers a claim is the same ad the team was paid to make. A generated video that is disclosed, substantiated and rights-cleared is harder to challenge and easier to cover; one shipped on hope is exposed on two fronts at once.

None of this means stopping the work. It means shipping generated video with the same discipline the rest of the craft already demands. Read the renewal, document the pipeline, close the rights, and treat affirmative AI coverage as a line item to discuss with the broker. The gap is real and widening, but it is also now something a prepared team can see coming and plan around.

Before a generated face fronts the brand, clearing a synthetic performer is the packet that proves origin, consent and provenance, and it is the evidence a standalone AI liability policy will ask to see.