An insurance policy document with a red EXCLUDED stamp overlaid on architectural blueprints generated by AI software
Policy & Regulation

Your Builder Used AI to Design Your Addition. The Insurer Just Excluded It From Coverage.

By Catherine Chen • July 5, 2026
← Back to all stories

On January 1, 2026, the Insurance Services Office published a new optional endorsement for commercial general liability policies. Its designation: CG 40 47. Twenty-three words in its operative clause, and it permits insurers to deny coverage for any claim arising from generative artificial intelligence.

If you build homes for a living, and you use AI tools anywhere in your workflow, this is the most important insurance document you have never read.

What CG 40 47 Actually Says

ISO defines "generative artificial intelligence" as "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." That definition sweeps in everything from ChatGPT-generated specifications to AI-assisted energy models to automated quantity takeoff tools. It applies to both Coverage A (bodily injury and property damage) and Coverage B (personal and advertising injury). There is no carve-out for professional review, no safe harbor for tools validated by a licensed engineer, no distinction between a chatbot hallucination and a purpose-built design platform.

ISO forms underpin approximately 82% of U.S. property and casualty insurance policies, and while CG 40 47 is optional, the construction law firm Cohen Seglias noted in April 2026 that optional ISO endorsements have a way of becoming standard at renewal.

82%
of U.S. property and casualty policies use ISO forms that now offer an AI exclusion endorsement

The Carriers Are Already Moving

Philadelphia Insurance and Hamilton Select have excluded AI-related claims from errors-and-omissions coverage entirely. AIG, Great American, and W.R. Berkley have filed for regulatory approval to introduce comparable exclusions, according to industry reports published in January 2026. Travis Landers, president of Risk Specialty Group, has reviewed dozens of design firm renewals over the past year. His assessment: "Only a handful so far carry absolute AI exclusions, but the rate of adoption is accelerating with each renewal cycle."

Translation: if you renewed your policy in 2025, you are probably still covered. If you renew in late 2026 or 2027, read the endorsement schedule before you sign. Those changes are not in the marketing materials. They are in the definitions section, where "professional services" gets quietly narrowed, or in an endorsement schedule your broker may not flag.

Your Contract Does Not Help

Standard AIA construction contracts were drafted for a world where licensed professionals make decisions and exercise judgment. A201-2017, the most widely used general conditions document in American construction, contains no provisions addressing AI use. It does not require disclosure of AI tools. It does not allocate liability for AI-generated errors. It does not define what level of human review satisfies the professional standard of care when AI produces the first draft.

Frantz Ward, a construction law firm, identified five specific gaps in current AIA contracts: standard of care ambiguity when AI is involved, undefined responsibility for AI-originated design errors, no intellectual property framework for AI-generated work, unallocated third-party technology risk, and misalignment between contractual liability and insurance coverage. Their conclusion was blunt: the party relying on AI to perform any obligation bears the risk if the AI produces inaccurate or incomplete information.

This pattern is consistent across jurisdictions, and it points to a surprisingly uniform legal consensus. A May 2026 joint analysis by construction law practitioners in England, Denmark, Mexico, and the United States reached the same finding: AI is treated as a tool, not an autonomous decision-maker, and the professional who deploys it remains liable for its output. Danish courts will assess "whether the professional acted prudently, understood the system's limitations, and appropriately verified AI-generated outputs." Mexican courts will apply "traditional regimes on breach, latent defects and professional negligence, generally placing risk on the party deploying the technology."

So the legal standard has not changed. You are still liable for everything you submit, every drawing, every calculation, every specification. But the insurance backing that liability may have vanished at your last renewal.

The Residential Builder's Blind Spot

ServiceTitan's 2026 Residential State of the Trades Report, a survey of 1,000 residential contractors, found that 74% view AI as an efficiency engine. Only about 25% are currently using it, but 73% believe starting early creates a competitive advantage, and 48% of early adopters report increased productivity. The pressure to adopt is real, and it will intensify as competitors who use AI underbid those who do not.

Meanwhile, U.S. homebuilders paid $1.071 billion in warranty claims in 2024, according to Warranty Week's analysis of 27 publicly traded builders. Lennar alone paid $290 million in 2023, and D.R. Horton paid $127 million. These are the costs when things go wrong under the current regime, where insurance generally covers defect-related claims.

Now imagine a fraction of those claims involve AI-generated design documents, energy calculations, or scheduling that contributed to a defect. Under the old policy, covered. Under a policy with CG 40 47 attached, potentially excluded.

$1.07B
paid in warranty claims by U.S. homebuilders in 2024 (27 publicly traded builders, Warranty Week)

Nobody has published data on what percentage of residential construction claims now involve AI-assisted work. That number does not exist yet, and nobody is counting. And that absence is exactly why insurers are moving preemptively: they cannot price AI risk because there is no actuarial history, so the rational response from a carrier's perspective is to exclude the unknown risk entirely and let the market figure out whether AI-generated design errors are really as dangerous as a hallucinated beam specification suggests they might be.

What a Builder Should Do on Monday

Risk Specialty Group recommends a four-step policy audit for every design or construction firm preparing for renewal:

First, pull your declarations page and read every endorsement schedule attached to your CGL and E&O policies. If an endorsement number starts with CG 40, read it twice.

Second, search the full policy text for the word "artificial." If it appears anywhere you did not expect, call your broker before you call your lawyer.

Third, check the definitions section. If "professional services" has been narrowed, or if there is new language around "technology-assisted" or "algorithm-generated" work product, the scope of your coverage may have changed without your E&O premium changing with it.

Fourth, verify your retroactive date. Claims-made policies only cover work performed after the retroactive date. If that date resets at renewal (which some carriers attempt during endorsement changes), you could lose coverage for AI-assisted work you performed last year.

Beyond the policy audit, construction attorneys are increasingly recommending bespoke contract amendments that address AI use explicitly: mandatory disclosure of AI tools in the project, defined human oversight and verification requirements, clear allocation of responsibility for AI-related errors, and coordination between contractual liability and available insurance coverage.

What This Does Not Prove

This analysis has limitations that matter. No residential construction defect claim has been litigated under a CG 40 47 exclusion. CG 40 47 is only six months old, and it is possible that carriers will adopt it slowly, that brokers will negotiate it out of renewal terms, or that state insurance regulators will reject it in jurisdictions with strong consumer protections. Residential construction, with its smaller project values and tighter margins, may not see the same adoption pace as the commercial sector where design-build firms carry $5 million E&O policies.

It is also true that most AI tools currently used in residential construction are supplements to professional judgment, not replacements for it. An architect who uses AI to generate initial design options but reviews and stamps every drawing still meets the standard of care. Whether an insurer would invoke the exclusion against that architect anyway, in an attempt to deny a claim, is an untested question.

The strongest argument against alarm is simple: the market may self-correct. If AI exclusions become standard and builders stop using AI tools, carriers lose the premium leverage that AI adoption creates. Specialty insurers may step in to offer AI-inclusive coverage at higher premiums, creating a new product line. The equilibrium may look less like a coverage crisis and more like a price adjustment.

But price adjustments in construction insurance flow downstream to homeowners. And right now, nobody is telling the person writing a check for a $600,000 custom home that their builder's insurance may not cover defects if the AI got the energy model wrong.

The form number is CG 40 47. Ask your builder if they have heard of it.

← Back to all stories