Your Builder's Insurance Covers Fire, Flood, and Collapse. It No Longer Covers the AI He Used to Design Your House.

On January 1, 2026, a four-page insurance form called CG 40 47 became available to every commercial general liability carrier in the country. It is an optional endorsement from the Insurance Services Office, written in the kind of dry actuarial language that nobody reads until a claim gets denied. What it does is straightforward: it excludes coverage for any loss "arising out of generative artificial intelligence."

That matters because ISO forms underpin roughly 82% of U.S. property and casualty insurance policies, according to Jones Day's analysis of the insurance landscape. That endorsement is optional today, but the 18 largest U.S. liability insurers have already filed for regulatory approval to exclude AI-related claims, with approval rates running between 80% and 100%. Major carriers like AIG, Great American, and WR Berkley aren't waiting for industry consensus. They're shipping their own exclusions.

Meanwhile, one in four residential contractors is now using AI in a meaningful way. That number comes from ServiceTitan's 2026 Residential State of the Trades report, which surveyed more than 1,000 contractors. A third of architecture firms incorporate AI tools into daily operations, according to American Institute of Architects research, and in large firms that number hits 61%.

So here is the collision nobody is talking about: the tools your builder uses to estimate your project, check your plans against code, and calculate your structural loads are increasingly powered by AI. And the insurance policy that's supposed to protect both of you when those calculations are wrong is increasingly designed to exclude AI from coverage. Those two trend lines crossed sometime this spring, and neither you nor your builder probably noticed.

What the Exclusion Actually Says

ISO defines generative AI broadly in the endorsement: "a machine-based learning system or model that is trained on data with the ability to create content or responses, including but not limited to text, images, audio, video, or code." Read that again. It covers essentially every AI tool a builder or architect might touch, from ChatGPT helping draft a project schedule to a specialized structural analysis tool running calculations through a machine learning model.

It applies to both Coverage A (bodily injury and property damage) and Coverage B (personal and advertising injury). In plain terms, if your beam fails because an AI tool hallucinated a load calculation, your builder's CGL policy may now deny the claim. If an AI-generated specification called out a material that doesn't meet code and the wall assembly fails a fire test, the insurer has language to walk away. The harm is physical. The AI was in the chain. The exclusion kicks in.

Berkley Insurance went further with what it calls an "Artificial Intelligence Exclusion (Absolute)" for directors and officers and errors-and-omissions policies. Their language doesn't just cover generative AI; it targets any loss "based on, arising out of, or attributable to" the use of AI, full stop. An architect who runs a quick code check through an AI compliance tool and misses a violation could find their professional liability carrier pointing to that exclusion at the worst possible moment.

Error Rates Are the Whole Problem

Insurers aren't doing this because they're paranoid. They're doing the math.

Stanford Law School research found that general-purpose AI tools produce hallucinations at rates between 58% and 88%. Even specialized "legal-grade" tools designed for accuracy show error rates of 20% to 33%. In the construction context, attorneys Richard Glucksman and Daniel Cribbs at Segal McCambridge laid out what those errors look like on a job site: skewed plans, incorrect specifications, contract risks, cost miscalculations, scheduling discrepancies. AI tools "rely entirely on the data they source," they wrote, "and if the information they pull is incomplete, outdated or biased, the answers will be inaccurate and often do not reflect field conditions."

Construction is not a domain where a 20% error rate is tolerable. A 20% error rate in structural calculations means one in five beams is undersized. That is not an insurance risk. That is a liability catastrophe, and insurers priced it accordingly by refusing to cover it at all.

Who Gets Hurt

Not the big firms. They have risk managers, in-house counsel, and the leverage to negotiate exclusions out of their policies or purchase standalone AI coverage. Jones Day's analysis explicitly advises policyholders to "negotiate to eliminate [AI exclusions] entirely or narrow their scope" and to seek "carve-outs for incidental use." That is sound advice for a firm with a $50 million policy and an insurance broker on retainer.

It is useless advice for a three-person residential framing crew in suburban Phoenix. Harvard's Joint Center for Housing Studies estimates that more than half of residential remodeling businesses with payrolls generate less than $250,000 in annual revenue. These are the contractors most likely to adopt AI tools because, as ServiceTitan's Angie Snow put it, "you're wearing a lot of hats" and AI can help you "streamline a lot of what you're doing." They are also the contractors least likely to read the endorsement schedule on their CGL renewal, notice the new AI exclusion, and have the bargaining power to remove it.

You, as a homeowner, sit at the end of this chain with the least visibility and the most at stake. You hire a contractor. He uses AI to generate your cost estimate. His architect used AI to check your plans against local code. Your structural engineer ran load calculations through an AI-assisted tool. Something goes wrong eighteen months later and a retaining wall shifts. You file a claim. And somewhere in the discovery process, the insurer's attorney finds AI fingerprints on the calculation that determined the footer depth, and CG 40 47 gives them a path to deny coverage.

Nobody told you any of this when you signed the contract.

Everything in One Table

What You Should Actually Do

If you are a residential contractor using any AI tool for estimates, plans, code checks, or scheduling, call your insurance broker before your next renewal. Ask specifically whether your CGL policy now includes ISO endorsement CG 40 47 or any carrier-specific AI exclusion. If it does, ask what it would cost to remove it, or whether a standalone AI liability product exists. These products are emerging but still immature, so the answer may be "not yet" and the honest next step is documenting your AI usage protocols, establishing human review checkpoints, and keeping logs of AI inputs and outputs so you can demonstrate due diligence if a claim ever lands.

If you are a homeowner about to start a project, ask your contractor directly: do you use AI tools for estimating, design, or code compliance? If yes, does your insurance policy cover errors from AI-assisted work? Most contractors will not know the answer, and that itself is information worth having before you sign.

If you are an architect or engineer, the AIA Trust's February 2025 guidance already flagged multiple AI risk categories including competence questions when relying on AI outputs, confidentiality breaches when uploading client data to AI platforms, and standard-of-care implications as AI use becomes common. Now add insurance coverage to that list: Berkley's absolute exclusion on E&O policies means your professional liability coverage may no longer respond to claims involving AI-assisted work product. Review your policy language with counsel, not just your broker.

A Familiar Playbook

This has happened before. When cyber risks first emerged, traditional CGL policies covered data breaches and network failures by default because no one had thought to exclude them. Then insurers woke up to the exposure, introduced broad cyber exclusions, and a separate $14 billion cyber insurance market grew to fill the gap. We are watching the same playbook unfold with AI, except faster. CG 40 47 took effect on January 1. Carriers are already filing.

Construction will follow the same arc eventually, as standalone AI coverage products mature, premiums stabilize, and underwriters develop frameworks for assessing AI governance. But "eventually" could be three to five years, and in the meantime every residential project that touches AI tools sits in a coverage gray zone that most participants do not even realize exists.

Limitations

CG 40 47 is optional, and I cannot tell you how many policies have been renewed with the exclusion attached since January 1 because that data does not exist publicly. That 82% figure describes ISO form usage broadly, not adoption of this specific endorsement. Residential-specific data on AI insurance gaps is thin; most analysis focuses on commercial and multifamily construction where the dollar exposure is higher. Standalone AI coverage products are emerging but their pricing, availability, and actual claim performance remain unproven. If you are making decisions based on this article, get an actual insurance review from a licensed professional. I write about construction technology, not insurance law.