On January 1, 2026, Verisk released two endorsement forms that most architects have never heard of. Forms CG 40 47 and CG 40 48 give insurance carriers standardized language to exclude losses arising from generative artificial intelligence. Written for commercial general liability policies, the language is already being adapted across professional liability lines. By the time your design team's errors-and-omissions policy comes up for renewal, the coverage they carried when they started using AI may no longer exist.
Nobody sent a memo to the homeowner, and nobody will, because the insurance industry's disclosure obligations run to the policyholder, not to the person whose $600,000 project depends on that policy existing when something goes wrong.
Five Carriers Have Already Moved
Berkley Insurance Company now applies what the market calls an "absolute" artificial intelligence exclusion. It covers directors and officers, errors and omissions, and fiduciary liability products. The endorsement names specific tools: ChatGPT, Bard, Midjourney, and DALL-E. If your architect ran a structural calculation through an AI assistant, and that calculation contained an error that led to a defect in your home, Berkley's position is that the resulting claim falls outside coverage.
Hamilton Insurance Group has its own generative AI exclusion, and Philadelphia Insurance together with Hamilton Select have gone further, stripping AI-related claims from E&O coverage entirely. AIG, Great American, and W.R. Berkley have filed for regulatory approval to introduce comparable exclusions, according to Risk Specialty Group, a Houston-based insurance and risk management firm that has reviewed dozens of design firm renewals since January.
Travis Landers, president and founder of Risk Specialty Group, reports that only a handful of current policies carry absolute AI exclusions so far. But the adoption rate is accelerating with each renewal cycle. By mid-2027, Landers expects the majority of E&O products to include some form of AI-related language, whether as a full exclusion or a narrowed definition of covered professional services.
Why Insurers Are Running
Stanford Law School research documents hallucination rates between 58 and 88 percent on general-purpose AI tools, and 20 to 33 percent on specialized systems designed for professional use. For a design firm, a hallucinated beam specification is not an abstract concern. It is a structural defect waiting for a rainstorm, a load calculation, or an inspection that never catches it because the inspector trusted the licensed professional who trusted the software that fabricated a code reference.
Insurers price risk based on predictability, and a licensed architect making a human error fits neatly into actuarial models refined over decades of claims data and loss ratios that carriers understand down to the basis point. An architect relying on a tool that generates plausible-sounding nonsense 20 to 88 percent of the time, depending on the task and the model, breaks every assumption those models rest on. From the carrier's perspective, covering AI-assisted design work is like insuring a driver who occasionally hallucinates traffic signals: you cannot price what you cannot model, so you exclude it and wait for someone else to figure out the math.
An ASCE survey from December 2025 found that only 27 percent of architecture, engineering, and construction professionals currently use AI in their work. But 94 percent plan to increase AI adoption in 2026. Carriers are not reacting to current claims. They are front-running a wave of adoption that has barely produced litigation yet, positioning exclusions before the first major AI-design-defect lawsuit sets a precedent that makes coverage retrospectively untenable.
Where This Leaves You
Say you commission a $600,000 custom home. Your architect's firm uses an AI-powered tool to run code compliance checks on the structural drawings, a workflow that is becoming routine at firms of all sizes, to the point where a CoLab survey of 250 engineering leaders found that 95 percent expect AI to triple design review speeds within 12 to 24 months and virtually none of those leaders said they planned to disclose the shift to clients. Your architect is probably not telling you they ran the plans through AI before stamping them. No state requires it. And as of the firm's most recent policy renewal, their E&O carrier may have excluded any claim arising from that AI-assisted review.
If the AI missed a code violation, and that violation caused a structural defect, and you sued the architect, here is what happens: the architect's carrier denies coverage under the new AI exclusion endorsement, your claim is valid, you may even win in court, and then collecting becomes a different problem entirely because you are now pursuing a judgment against the firm's assets rather than against a million-dollar professional liability policy that was designed for exactly this scenario.
Small firms are especially exposed. A five-person architecture practice carrying a $1 million E&O policy is one AI exclusion away from being effectively uninsured for the type of work that increasingly defines how they operate, and if the firm lacks the assets to satisfy a judgment, the homeowner absorbs the loss, which means you are funding the gap between what the architect's technology promises and what the architect's insurance actually covers.
Contracts Haven't Caught Up Either
Standard form agreements from the American Institute of Architects and ConsensusDocs were written before AI entered design workflows, and they do not address who bears responsibility when an AI tool produces an incorrect output that a licensed professional incorporates into a design, nor do they require disclosure of AI use or allocate liability between the design firm, the AI vendor, and the contractor who builds from AI-assisted plans.
Kutepova's analysis identifies five gaps that existing contracts fail to cover: selection and configuration of AI tools, ownership of AI-generated data and designs, liability for autonomous system failures, whether AI outages constitute force majeure, and audit rights over AI decision logs in the event of a dispute. Without express contract language addressing these questions, outcomes depend on litigation where neither side has clear precedent to cite.
Bilzin Sumberg's November 2025 analysis for Law360 catalogs the full scope: contractual liability for AI-generated errors, intellectual property disputes over AI training data that may contain copyrighted works, data privacy obligations when AI platforms process project-specific financial and geolocation data, and consumer protection exposure when builders and architects fail to disclose AI use to buyers.
What to Do Before You Sign a Contract
If you are commissioning design work for a residential project in 2026, five questions belong in your first conversation with the architect.
First: Does your firm use AI tools at any point in the design, code compliance, or specification process? Expect a yes, and if the answer is no, verify it independently, because a survey by Chaos of 1,200-plus architects found AI adoption growing across firms of every size.
Second: Does your current E&O policy contain an AI exclusion? Risk Specialty Group recommends a four-step audit: check the declarations page for endorsement schedules, search the policy body for any reference to "artificial," review the definitions section for narrowed "professional services" language, and verify the retroactive date that governs claims-made coverage for past AI-assisted work.
Third: Will the contract include AI-specific liability allocation? You want clauses that define which party bears responsibility for AI tool selection, require verification by a licensed professional before any AI output is incorporated, and specify indemnification if AI-driven recommendations produce defects.
Fourth: What verification process does the firm use for AI-generated outputs? A licensed engineer reviewing and stamping AI-assisted calculations is different from a junior designer copying ChatGPT's structural suggestion into Revit. Some carriers have shown willingness to negotiate carve-backs for supervised AI use when a firm can document tool approval, verification steps, client disclosure, and licensed professional sign-off.
Fifth: Will the firm disclose AI use to me in writing? No state currently mandates this for design services. Contractual disclosure requirements are the homeowner's only mechanism to establish a record of what tools were used and when.
Strongest Counterargument
Insurance exclusions are not permanent. They are pricing signals. When carriers excluded cyber risk in the early 2010s, that exclusion eventually produced a standalone cyber insurance market worth $14 billion by 2025. AI design coverage may follow the same trajectory: a period of exclusion, followed by specialized products that price AI risk explicitly rather than bundling it into general E&O. Early signs exist. Risk Specialty Group reports that some carriers will negotiate carve-backs today for firms with documented AI governance programs. Within two to three years, standalone AI professional liability products may emerge from the surplus lines market.
That is a reasonable forecast, and it may even prove correct. It is also irrelevant to anyone signing a design contract this year. Standalone AI coverage does not exist yet. Your architect's policy renewed in March, and if the carrier added an AI exclusion endorsement, the coverage gap is live right now, not in some future market correction where specialized products arrive to fill the hole that carriers spent 2026 deliberately creating. You cannot insure a 2026 project with a 2029 product.
Limitations
This analysis relies on publicly reported carrier actions and the assessments of one insurance advisory firm, Risk Specialty Group, that has a commercial interest in selling risk management services. We could not independently verify Landers's claim about the number of renewals containing AI exclusions, because policy endorsement data is proprietary and carriers do not publish adoption statistics for specific exclusion forms. Stanford's hallucination rate figures (58 to 88 percent for general-purpose tools, 20 to 33 percent for specialized systems) apply to legal research tasks, not architectural design specifically. Hallucination rates in structural engineering and code compliance tools may differ, though no published study has measured them in that context. Adoption rates vary by survey methodology: Research.com reports 73 percent for U.S. architecture firms, RIBA reports 59 percent globally, and ASCE reports 27 percent across the broader AEC sector. We cannot reconcile these figures because the surveys use different definitions of "using AI" and different sample populations. No AI-design-defect lawsuit has reached a published verdict in the United States as of May 2026, which means the legal theories described here are untested in court.