If you used an AI tool to write one estimate this year, part of your general liability coverage may already have an expiration date on it. Not because you did anything wrong — because the policy language changed underneath you this year, and most contractors haven’t read the new pages yet.
The Insurance Services Office — whose forms underpin the large majority of U.S. commercial policies — introduced three optional generative-AI exclusion endorsements with a January 2026 edition date: CG 40 47, CG 40 48, and CG 35 08. They give your carrier the option to carve claims arising from generative AI straight out of your commercial general liability coverage. The forms aren’t automatic — adoption varies by carrier, state, and renewal date. But several major carriers have started attaching them at 2026 renewals, and one analysis counted more than 4,000 supporting filings, most touching the exact lines contractors carry: CGL, umbrella and excess, commercial package, businessowners, and E&O. And unlike the cyber transition, there’s no admitted standalone product that buys the excluded exposure back — the gap is just a gap.
Before these endorsements, most standard CGL policies didn’t mention generative AI at all. The industry is treating it as “silent cyber” all over again — coverage that existed because nobody had named the exposure. The new forms name it, then carve it out.
The three forms, and the one that should worry you most
The broadest form, CG 40 47, strips out bodily injury, property damage, and advertising injury claims arising from generative AI. CG 40 48 is narrower — advertising injury only. And CG 35 08 targets completed operations: the coverage a contractor relies on after a job closes, when the real claims show up.
That last one deserves a second look. Your work lives in completed operations. The water’s extracted, the structure’s dry, the build-back is done — and six months later someone alleges the drying plan was wrong. If any piece of that job’s documentation trail runs through a generative-AI tool, CG 35 08 gives the carrier an argument that the claim falls in the carve-out.
The trigger is smaller than you think
Under the broadest form, a claim need only arise out of generative AI. Not “be primarily caused by.” Not “be mostly AI’s fault.” Just connected to it.
“A contractor might use AI for one estimate, takeoff or design decision and do the rest of the job without it,” Matt Levin, head of research at Trades Coverage, told Insurance Business. “If that work contributes to a later claim, the exclusion could apply even though AI was used for only a small part of the project.”
Read that again. One AI-assisted estimate on an otherwise conventional job. That’s the exposure. And contractor AI adoption is accelerating into exactly this gap — ServiceTitan’s 2026 Commercial Specialty Contractor Industry Report found 38% of contractors now report measurable business impact from AI, up from 17% a year earlier, with cost estimating and bid management the most common uses. The industry is running toward AI with its coverage quietly running the other way.
Your AI paper trail just got more readable
The coverage change arrived alongside a courtroom development worth your attention. In February, a federal judge in New York issued the first ruling in the country on whether AI-assisted work can be protected by attorney-client privilege — United States v. Heppner in the Southern District of New York. The court found the specific AI chats at issue weren’t privileged on those facts. But its reasoning suggested AI tools genuinely directed by counsel could be treated differently — an analogy to the long-standing Kovel doctrine that protects work by non-lawyer experts acting on a lawyer’s behalf.
The practical point isn’t the doctrine. It’s what the ruling assumes about your records: that they exist, in volume, and that they’re readable. A new Reed Smith / Haven Safety AI whitepaper puts it bluntly. AI doesn’t create some new theory of liability. Courts have always judged organizations by what they knew and what they did about it. AI just makes that record bigger, more structured, and easier to read — in litigation, in regulatory inquiries, and in insurance claims.
Translate that to a restoration shop. Your AI-generated moisture maps, your auto-drafted drying logs, your AI-annotated photo documentation, the scope narrative the tool helped write — that’s now a cleaner, more searchable record of what you knew and when you knew it. That’s not an argument against using the tools. It’s an argument for governing them like the evidence they’ll become.
The two-lane fix
The whitepaper’s answer is a two-lane operating model, and it ports neatly to a contractor’s world:
Lane one — routine operations, AI moves fast. Estimates, documentation, moisture readings, daily logs. Outputs here are ordinary business records. Use the tools freely, but treat everything they produce as something a claims adjuster or an opposing attorney will eventually read. Because they will.
Lane two — sensitive matters, counsel directs. The job that’s heading toward dispute. The incident with injury potential. The file where coverage is already in question. Here, AI work happens under counsel’s direction, access is restricted, distribution is minimized, and workspaces are kept separate — the structure that gives privilege its best chance under Heppner.
And across both lanes: keep a hard line between draft AI outputs and approved conclusions. A draft hypothesis the tool floated at 9 a.m. should never be discoverable as your company’s finding at 9 p.m. Label them, segregate them, and make sure the approved record shows a disciplined process — clear findings, clear disposition of every recommendation, no unresolved AI suggestions sitting in the file.
As Reed Smith partner John Ellison put it: this isn’t really an AI problem. “It’s a documentation and governance problem that AI happens to be accelerating.”
What to do Monday morning
- Pull your renewal paperwork. Look for CG 40 47, CG 40 48, or CG 35 08 — or any generative-AI exclusion language your carrier wrote itself.
- Call your broker, don’t email. Ask directly: did my CGL pick up an AI exclusion at renewal? What’s my completed-operations exposure if it did? This is a five-minute conversation that beats a coverage fight later.
- Inventory where AI touches your jobs. Estimating, documentation, photo annotation, customer communication — write the list down. You can’t govern what you haven’t mapped, and the “arising out of” trigger means even incidental use counts.
- Separate drafts from findings. Today. Whatever tools your team uses, make sure AI-generated drafts are labeled and segregated from approved conclusions before they land in the job file.
- For the sensitive file, get counsel in before the AI does the work — not after. Privilege is a structure you build in advance, not a label you apply afterward.
The industry spent a decade learning that the camera phone made every jobsite a record. AI is doing the same thing to the office side of the job — and now the policy language is catching up. The contractors who treat their AI output like the evidence it is will be fine. The ones who never read the new pages and never labeled their drafts are the ones the exclusions were written for.
This article is informational, not legal advice. Coverage turns on your specific policy language — read it, and talk to your broker and your attorney.
Sources: The AI clause that may already be on your contractor client’s policy (Insurance Business); New 2026 insurance exclusions and first-of-its-kind federal ruling raise the stakes for AI-assisted safety investigations (ACCESS Newswire via Morningstar), reporting on the Reed Smith / Haven Safety AI whitepaper.