Who Owns the Mistake?
AI in Architectural Practice
by Trent Cotney, Partner, Adams & Reese, LLP
(Editor’s Note: Trent Cotney, partner at Adams & Reese, LLP, is dedicated to representing the roofing and construction industries. Cotney is General Counsel for the Western States Roofing Contractors Association and several other industry associations. For more information, contact him at (866) 303-5868 or go to www.adamsandreese.com.)
Artificial intelligence (AI) is no longer a future issue for architects. It is already being used to summarize meetings, generate concepts, assist with specifications, research code provisions, review documents, create images, and accelerate repetitive design tasks. Technology offers real efficiencies, but it also creates a basic risk-management question: when AI gets something wrong, who is responsible?
For architects, AI is a tool, not a licensed design professional. An architect who incorporates AI-assisted work into a project still remains responsible for the professional services the architect provides. That principle matters because AI can produce information that looks authoritative even when it is inaccurate, incomplete, outdated, or entirely fabricated. A convincing answer is not necessarily a correct answer.
This risk becomes particularly significant when architects use generative AI for code research. Building codes are highly technical, frequently amended, and often supplemented by state or local requirements. An AI platform may cite the wrong edition, miss a local amendment, confuse jurisdictions, or generate a code section that does not exist. Architects should treat AI-generated code analysis as a research starting point, not as a substitute for reviewing the actual governing code and applicable amendments.
The same concern applies to specifications and construction documents. AI may help draft language or identify inconsistencies, but it does not understand the complete project in the same way the design team does. A specification generated from incomplete prompts may conflict with drawings, manufacturer requirements, testing standards, or other contract documents. If that error reaches the field, the fact that AI generated the language will provide little comfort when the parties begin allocating responsibility for additional cost or delay.
Confidentiality presents another concern. Architects routinely possess sensitive owner information, proprietary designs, budgets, security information, and other project data. Firms should know what happens to information entered into an AI platform before employees upload project documents or client information. The terms governing an enterprise AI platform may differ significantly from those governing a free public version. A firm that has not established rules for what employees may upload creates unnecessary exposure.
Copyright is also developing quickly. The United States Copyright Office has made clear that copyright protection continues to depend on human authorship. AI-assisted work may qualify for protection when a human author contributes sufficient creative expression, but purely AI-generated material presents different issues. Architects should also consider the source material used to create prompts and outputs, particularly when generating renderings, images, design concepts, or other creative content. The convenience of producing an image in seconds does not eliminate intellectual-property concerns.
The standard of care remains the central issue. Architects are judged by the professional services they provide, not by the sophistication of the software they use. A firm cannot reasonably argue that an error should be excused because an algorithm created it. If anything, the increasing availability of AI may eventually affect expectations concerning document review, coordination, and quality control. The technology may change how architects perform their work, but it does not eliminate professional judgment. That verification process should be documented when AI materially contributes to technical analysis, specifications, or other project decisions.
Architectural firms should therefore establish an AI policy before widespread use develops informally. At a minimum, firms should identify approved platforms, prohibit entry of confidential information into unapproved systems, require independent verification of technical and code-related outputs, address copyright and ownership concerns, and require meaningful human review before AI-generated material becomes part of a project deliverable. Firms should also train employees on the limitations of these systems. A written policy that no one understands or follows provides little protection.
Contract language deserves attention as well. Architects should review agreements to determine whether existing provisions adequately address technology use, ownership of deliverables, confidentiality, and limitations of liability. Owners may also begin asking whether and how AI will be used on their projects. Transparency can reduce misunderstandings, but architects should avoid contractual language that guarantees the accuracy or performance of AI systems they do not control.
AI will continue to become part of architectural practice because the efficiency gains are too significant to ignore. The better approach is not to prohibit the technology, but to control its use. Architects should view AI the same way they view any other powerful professional tool: understand its capabilities, recognize its limitations, verify its work, and never substitute automation for professional judgment. When the final documents leave the architect’s office, responsibility will still rest with the professionals who decided they were ready to issue.
