The AI Mistake Your Policy Won't Cover

A client called me a few weeks ago about his renewal packet. Not about the premium. About one page in it he could not make sense of.

It was a single-page endorsement, and about a third of it was a definition of the phrase "generative artificial intelligence." He runs a specialty contracting business in North Jersey, twenty-two people, and he wanted to know whether the page meant anything or whether it was boilerplate.

I told him I am not an insurance guy and he should ask his broker. Then I read the page, and called him back to say he should ask his broker this week.

Something changed in January and nobody sent a memo

For the last few years, most business insurance treated AI roughly the way it treated email in 1998. It did not mention it. That silence generally worked in your favor, because if a policy does not exclude something, there is at least an argument that it responds. The industry has a name for this arrangement: silent AI.

Silent AI is ending.

In January, ISO, the organization that writes the standard forms most carriers build their policies from, released three generative AI exclusions for commercial general liability. Verisk, ISO's parent company, describes them without much ceremony. CG 40 47 excludes bodily injury, property damage, and personal and advertising injury arising out of generative AI. CG 40 48 does the same for personal and advertising injury alone. CG 35 08 covers products and completed operations. Each one carries its own definition of generative artificial intelligence.

Attorneys who work in this space say carriers have been filing with state regulators for permission to use those three forms. Alana McMullin, a partner at Lathrop GPM who defends insurance disputes, told Claims Journal in July that the ISO forms "were the spark of this AI exclusion boom."

She was also careful about what nobody knows yet. Insurers are still deciding how far to take this and which lines it lands on, and some may price the risk rather than exclude it, because a policy that excludes something every business now does is a harder policy to sell. Her read was that renewals are where this becomes visible.

Not a headline. A renewal packet.

There is also a broader version already in the market. Berkley introduced what is being called an absolute AI exclusion for directors and officers, errors and omissions, and fiduciary liability products. It reaches the use or development of AI, content generated with AI, a failure to identify AI content that came from a third party, and even a company's own policies and procedures about AI use.

Notice the last item. It is not only what your AI did. It is what your rules about AI did or did not say.

You do not have to be an AI company to have AI exposure

This is where most owners tune out, and it is exactly the wrong place to tune out.

Fenwick's insurance practice made the point clearly in June: a retailer, a manufacturer, a professional services firm, none of them AI companies, all of them creating AI exposure the moment an employee uses a third-party tool to draft code, summarize a contract, screen applicants, analyze customer data, or write marketing copy.

Their second point is sharper. Coverage usually turns on how a claim is characterized, not on what technology was involved. An AI-assisted hiring screen that draws a discrimination claim is an employment claim. AI-written marketing copy that lifts someone's material is an intellectual property claim. Somebody pasting a client list into a consumer AI tool is a privacy claim. Same tool, three different policies, three different exclusions, and no guarantee any of them line up.

Fenwick also flagged that the narrowing is often not one obvious exclusion stapled to the front. It shows up in revised base forms, tightened definitions, new application questions, and carve-backs buried in endorsements. You can pass right over it while checking that your limits did not change.

The other side of the ledger is moving too

Coverage narrowing would matter less if claims were flat. They are not.

A Gallagher study cited by Claims Journal counted a 978% increase in AI-related lawsuits from 2021 to 2025, and a 137% increase from 2024 to 2025 alone. Patent claims, copyright claims, and privacy claims made up the largest slices.

And last week Insurance Journal published something that should get your attention even if you never touch an AI tool. Plaintiffs' firms are now running online mock juries of up to 1,200 people, mining data on jury pools, and using AI to test arguments before trial. The article cites a Marathon Strategies report finding that verdicts of at least $10 million jumped 41% from 2024 to 2025, the highest level since 2009.

So the people who might sue you are getting better tools, and the page behind you is getting thinner. Those two trends arrived in the same year.

Four questions, one afternoon

I am not going to tell you what coverage to buy. That is your broker's job and it depends on facts I do not have. But these are the questions I would want answered in writing.

Does any policy in my program exclude or limit AI-related claims, and on which lines? Ask about general liability, E&O, D&O, cyber, and employment practices separately, because they are moving independently.

How is AI defined in that endorsement? The definition is the whole ballgame. A narrow one aimed at generative tools is a different animal than one that sweeps in any automated decision system, which might include software you have run for years without thinking of it as AI.

Can the coverage be written back? Affirmative AI endorsements and standalone products exist. They are early and not widely adopted, but the answer is not automatically no.

And then one for yourself: where does my business actually use AI? Write it down. Every tool, who uses it, and for what. You cannot get a straight answer from a broker about exposure you cannot describe, and application questions are getting more specific every cycle. An inaccurate answer on a form is its own problem later.

The cheapest coverage is not needing it

None of this is a reason to pull back from AI. The businesses I work with are getting real time back, and that does not reverse because a form changed.

It is a reason to keep a person in the loop on anything that leaves your building, and to stop letting a tool make the final call on hiring, credit, or safety. Those were good habits when the policy was silent. They are load-bearing now.

The contractor made the call. Nothing bad has happened to him and probably never will. He just knows where he stands now, which is more than he knew when he opened the envelope.

If you want help putting together an honest inventory of where AI actually touches your business, that is something we do at Nexera Intelligence. Bring the renewal packet too.