Your Built Environment Project Uses AI: Does Your Risk Allocation Reflect That?

Artificial intelligence is no longer an emerging technology in the design, engineering and construction industries – it is already doing substantive project work, often without anyone stopping to ask what that means for liability.

Two inconsistent facts frame the current landscape:

1. A large majority of U.S. commercial property and casualty policies are underwritten on ISO/Verisk forms that, as of January 1, 2026, may carry standardized generative-AI exclusions.
2. Conversely, none of the construction industry’s most widely used contract templates contain a clause that addresses artificial intelligence at all.

On any given project today, AI tools are generating structural calculations, flagging safety violations and producing cost estimates that feed directly into deliverables – while the insurance backing that work and the contracts allocating responsibility for it were both written for a world in which AI played no role.

AI Is Already Producing Work Product on Built Environment Projects

AI tools are no longer confined to back-office efficiency gains – they are generating the substantive content of project deliverables across almost every discipline.

• In architecture and design, for example:
  • Generative space-planning tools may now lay out commercial office and kitchen floor plans;
  • Sketch-to-render programs can turn a rough hand-drawn sketch into a client-ready 3D visualization; and
  • Material-selection tools recommend building products based on climate, cost and sustainability inputs. These outputs can move directly into a construction document.
• In engineering, generative structural-design tools may calculate configurations to meet strength and load requirements, and simulation tools run live stress and thermal testing. These are not drafting aids – they are producing the calculations a licensed engineer is professionally and legally responsible for.
• In construction, scheduling and jobsite-monitoring platforms increasingly rely on computer vision to flag safety violations in real time.
• In project management, AI is being used to assess labor-demand simulations against deadlines, analyze cash flow against invoices and labor costs, and audit material spec sheets against actual purchases – decisions that used to sit squarely with a human project manager.
• On the intellectual property side, firms are using AI defensively – scanning public platforms for unauthorized use of proprietary design algorithms, while also using it to confirm that a new design does not infringe another firm’s protected building information model.

None of these AI use cases is speculative: they are running on active projects right now, frequently through tools whose AI functionality a firm may not have specifically negotiated or even fully identified.

Standard Indemnification Language Was Not Written for an AI-Caused Error

Consider the indemnification clause in the AIA A401 Standard Form of Agreement Between Contractor and Subcontractor. In substance, the language appears in some form across countless subcontractor agreements industrywide. Section 4.6.1 requires the subcontractor to indemnify the contractor, owner and design team for claims “to the extent caused by the negligent acts or omissions of the Subcontractor.”

What constitutes negligence is ambiguous with respect to the use of AI in project deliverables. For example:

• If a structural engineer runs a beam design through a generative structural analysis tool and the tool’s underlying algorithm miscalculates load capacity, has the engineer committed a “negligent act or omission”?
• What if the tool was required or recommended by the general contractor or the owner?
• What if the error originated in AI functionality embedded in a design or scheduling platform the firm licensed for an entirely different purpose and never separately vetted?

Standard indemnification language answers none of these questions, likely because it was drafted for a world in which every error traced back to an identifiable human decision. It does not say who owns a mistake that originates inside a black-box algorithm. Parties to a contract may be forced to litigate their way through the ambiguity instead of mitigating and allocating the risk through the contracting process.

The Contract Gap Is Now Colliding With an Active Insurance Gap

Insurance industry standard ISO Forms CG 40 47, CG 40 48 and CG 35 08 became available for insurers to attach to commercial general liability (CGL) policies renewing on or after January 1, 2026, and exclude bodily injury, property damage and products/completed-operations liability “arising out of, or attributable to” generative AI. Several insurers have followed with their own proprietary exclusion language barring claims tied to AI use “by any person or entity” (broad enough to reach AI embedded in vendor software a firm never knowingly activated).

The result: an architect’s E&O policy and a general contractor’s CGL policy can carry separate AI exclusions on the same project, leaving both firms uninsured for the same AI-generated error – whether in cost estimation, scheduling, code compliance or structural analysis – until a claim is denied.

Regulators Are Raising the Stakes for Contract Silence

Several developments now directly influence how construction contracts should be written.

• California’s new Civil Code Section 1714.6, effective January 1, 2026, bars a defendant who developed, modified or used an AI system from arguing that the AI “autonomously caused” the harm. This means an architect or engineer who uses an AI tool to calculate structural loads or manage a schedule cannot point to the tool itself as a defense if it fails – responsibility stays with the human user.
• California’s Civil Code Sections 3110 and 3111 (the AI Transparency Act) became operative for covered generative-AI providers on August 2, 2026, and require those providers to embed digital watermarks in AI-generated images, video and audio, which may be directly relevant to the architectural renderings, marketing visuals and project photography firms rely on in bidding, procurement and disputes.
• Additionally, at the federal level, the General Services Administration has proposed contract clauses that would require contractors to disclose when large language models process government project data, and to flow the same data safeguarding obligations down through every subcontractor tier on federal work (GSAR 552.239-7001). It remains in proposed rulemaking as of this writing, but firms performing public sector work should not wait for a final rule to start preparing.

Closing the Gap: What This Means for Your Contracts

At a minimum, contracts should:

• Define “artificial intelligence” and “AI-assisted work” in the agreement, expressly including AI embedded in vendor platforms, not only tools a firm affirmatively selects;
• Replace or supplement general negligence-based indemnification with language that expressly addresses AI-generated errors, hallucinations and model drift;¹
• Require documented human review before AI-assisted work enters a deliverable tied to a defined standard of review, avoiding both an unqualified accuracy warranty and a disclaimer that review cannot be relied on;
• Address who owns AI-generated content, what project data can be entered into third-party AI platforms and how long that data is retained;
• Build in ongoing AI-use disclosure obligations, and consider requiring adherence to the NIST AI Risk Management Framework or ISO/IEC 42001 certification for higher-risk engagements;² and
• Coordinate any new indemnification or warranty language with the firm’s current insurance program before finalizing it.

Looking Ahead

The gap between how AI is actually used on a project and how contracts and insurance respond may be widening. The ISO exclusion rollout and the emerging regulatory patchwork are moving faster than most standard contract forms. Firms that inventory their AI use, revise their templates deliberately rather than relying on forms that are silent on the issue, and confirm those contract terms are actually backed by available insurance coverage will be substantially better positioned when a dispute arises.

If your business operates in the built industry sector and you have questions about potential liability risks related to project contracts and/or insurance coverage, please contact Michael Gonzales, Richard McDonald or Alana McMullin, or your regular Lathrop GPM attorney.

¹ Model drift is the gradual decline in an AI tool’s accuracy or output quality as real-world conditions diverge from the data it was trained on.

² The NIST AI Risk Management Framework (AI RMF 1.0), released January 26, 2023, is voluntary, non-certifiable guidance built around four functions (Govern, Map, Measure and Manage) for identifying and managing AI-related risk across a system’s lifecycle. ISO/IEC 42001, published in December 2023, is the first international standard for AI management systems and, unlike the NIST framework, is certifiable: organizations can obtain third-party certification from an accredited body, similar to ISO 9001 or ISO 27001, through a staged audit process with three-year recertification cycles.