Enterprise Liability for (Some) AI: When the Enterprise Is the Cheapest Deterrable Cost Avoider

<p><span>Tort scholars have debated the concept of enterprise liability for decades: what it means, whether it is a socially desirable doctrine, and when it should apply and when not. This Article offers a new defense of a particular conception of enterprise liability. Specifically, it argues that enterprise liability is best understood as a regime that holds enterprises liable, without regard to fault, for the harms that their products, services, and activities cause when three conditions hold. The first is that the enterprise is the cheapest (or second cheapest) cost avoider for the class of harms at issue. The second is that the victims are systematically undeterrable, because they respond too weakly to the incentives a liability rule creates for it to change their behavior, whether on account of cognitive bias or because a first-party insurer absorbs the loss. The third is that the residual risk remaining after all cost-justified precautions is idiosyncratic and insurable. Conditions one and two together identify the enterprise as the cheapest cost avoider among the parties who will actually respond to liability, which this Article calls the cheapest deterrable cost avoider.</span></p> <p><span>This framework is especially important now. It is increasingly the case that many potentially harm-causing activities can be, and are being, handled by AI systems. AI can drive cars, diagnose diseases, manage investment portfolios, and draft legal documents. And all of those can go awry. Such situations present an especially strong case for enterprise liability, under the conditions just described. But it is important that any new enterprise liability regime apply not only to AI-enhanced activities, but to all activities that satisfy the conditions. A regime limited to AI would tax the AI version of an activity and not the human version, and so would discourage the substitution toward AI even where AI is the safer choice. A regime defined by the conditions applies on the same terms to the autonomous vehicle and the conventional car, and to the diagnostic algorithm and the unaided physician, so a consumer has no cost-based reason to choose the more dangerous option.</span></p> <p><span>The three conditions are typically satisfied for physical injury and property damage, where enterprise liability is therefore likely to be the optimal rule, and not for pure economic loss. For the pure economic loss that follows from financial or legal advice, with no physical injury behind it, the third condition fails. The residual risk of advice is the risk that markets will move or laws will change, which is systematic and cannot readily be insured. Enterprise liability should not extend to that residual risk. Advisors remain liable under existing tort and fiduciary law for negligent advice, for breaches of the duty of loyalty, and for execution failures, whether the advice comes from an AI system or a human.</span></p>