The rule of strict products liability in section 402A of the Restatement (Second) of Torts is widely assumed to have been formulated to address manufacturing defects. Relying on this assumption, the Restatement (Third) of Torts defines manufacturing defects as departures from the product’s design and therefore from the manufacturer’s manifest intent. Because virtually all product malfunctions stem from manufacturing defects, the Third Restatement extended this logic to define malfunctions as departures from manifest intent as well. This move effectively discards the section 402A consumer expectations test as a redundant relic from the early days of products liability.
This framework fails to anticipate a structural difference between malfunctions in physical goods and failures in the intangible software running artificial intelligence (AI) models and systems. For example, a manufacturer manifestly intends for an airbag to deploy according to discrete physical triggers. If the device meets those specifications but a passenger is still injured, that outcome is not contrary to the manufacturer’s manifest intentions and does not constitute a malfunction. AI safety measures have a fundamentally different logical structure. Because AI safety objectives are typically coded as preventative goals-“do not crash,” “filter harmful content”-the developer’s manifest intent is synonymous with the safe outcome itself. Every harm the safety measure was coded to prevent necessarily constitutes a departure from manifest intent and satisfies the Third Restatement’s definition of a malfunction, regardless of how carefully the system was designed or how unavoidable the failure. The absence of perfection triggers strict liability, penalizing safety measures designed to reduce harm-a perfection tax.
The original understanding of section 402A offers a way out. Building on Michael Green’s reexamination of the doctrine’s origins, this Article argues that section 402A’s true foundation lies in a tort-warranty theory centered on product malfunctions that frustrate consumer expectations. That tort warranty—a common-law doctrine not limited by privity, disclaimer, and notice requirements—is distinct from the contract warranty section 402A is conventionally said to have absorbed. When “defect” is defined as such a performance-based malfunction rather than a blueprint-based flaw, the ambiguities that have long vexed section 402A dissolve.
That foundation also resolves the perfection-tax problem. Under the tort-warranty standard, the operative question is not whether an AI safety guardrail departed from the developer’s manifest intent, but whether it performed more dangerously than consumers would expect. When informed by an adequate warning of inherent dangers, consumers do not expect the complete elimination of risk. The mere occurrence of an AI-caused harm is not necessarily a malfunction, so the perfection tax disappears. This restoration of section 402A’s original meaning does more than set the historical record straight. It is the condition under which tort law can encourage, rather than penalize, the safety measures on which AI’s responsible development depends.
Read the full article:
Geistfeld, Mark, Recovering Strict Products Liability for the Age of AI (July 24, 2026). NYU School of Law, Public Law Research Paper Forthcoming, NYU Law and Economics Research Paper Forthcoming, Available at: https://ssrn.com/abstract=7202358