Private Law Theory

ABSTRACT
In April 2026, the Sixth Circuit removed appointed counsel from an indigent defendant’s appeal and denied Criminal Justice Act compensation after the attorney admitted that CoCounsel, Thomson Reuters’s artificial intelligence (AI) legal research platform, fabricated quotations and misrepresented case holdings. The sanctioned attorney lost his fee. The defendant lost continuity of appellate counsel. Thomson Reuters, which had marketed CoCounsel with affirmative representations of reliability and whose platform generated the fabricated output, was not named in any proceeding. This asymmetry is the subject of this Article. This Article proposes the Architectural Defect Doctrine: a framework applying strict products liability to legal AI vendors whose deliberate choice to deploy stochastic generative architecture for high-stakes legal research tasks constitutes a design defect when a feasible zerofabrication alternative exists. Three independent theories support the doctrine. First, commercial legal AI tools are products under the Restatement (Third) of Torts: Products Liability under the Jeppesen mass-production-for-professional-reliance line and Garcia v Character Technologies, Inc. Second, vendors’ affirmative marketing representations regarding reliability create express warranty obligations under the Uniform Commercial Code (UCC) § 2-313 that disclaimers in terms of service cannot defeat under § 2-316(1). Third, the failure to warn theory operates independently: a vendor who knows its product’s documented hallucination rates and markets it as reliable has not adequately warned the professional users on whom it relies for revenue. An attorney’s failure to verify AI output is comparative fault under the Daly framework, not a superseding cause that extinguishes vendor liability. The learned intermediary defense fails where vendor marketing actively suppressed the attorney’s risk assessment. This Article predicts a Farris-pattern verdict within five years and provides the doctrinal framework for that case.

Austin Litle, Alexis and Morris, JD and Das, Deepankar, When the Tool Breaks: Product Liability and the Legal AI Vendor (May 9, 2026).