Alex Chan’s working paper, Optimal Medical Liability for AI, is dated 9 June 2026 in the Harvard Business School version and was circulated by NBER in June 2026. The paper develops a theoretical framework for situations where AI acts as a doctor rather than only assisting a clinician. It examines how the available medical record affects the incentives and responsibilities that a liability arrangement can create.

The argument is conditional: different information can support different rules. It is not a new law, a clinical safety finding or a recommendation that an employer select a particular liability regime. For benefits teams reviewing digital health suppliers, its relevance is the question of what evidence exists when a decision or service is disputed.

Understand the record behind the service This editor recommends asking a prospective provider what information its service records and how that information can be used to explain an outcome. The discussion can cover the inputs considered, the advice delivered, warnings, follow-up instructions and the responsibilities assigned to people and systems. A product demonstration should be connected with the actual service proposed in the contract.

The employer should also understand the provider’s process for responding to complaints or suspected errors. A named contact and a defined escalation route can make the operating relationship clearer. The existence of a log does not by itself establish that a complaint will be resolved or that the information is complete and intelligible.

Connect evidence with responsibilities Supplier due diligence can examine who delivers clinical care, who operates the technology and who addresses service failures. Those roles may involve different organisations. The employer needs an explanation of how they work together and which commitments it can rely on. A claim that a tool is advanced should not replace that account.

Access to records also raises practical questions about confidentiality and retention. Benefits teams should establish what information they need for programme oversight and what should remain with the appropriate clinical or service provider. The objective is a defined governance process, not unrestricted employer access to personal medical information.

The research offers a framework for asking how information supports accountability. Employers can translate that question into a more concrete supplier review while leaving legal and clinical determinations with the appropriate specialists. Clear responsibilities, usable explanations and a documented process for addressing problems are meaningful elements of that review.

Sources: Source de référence · Working paper HBS 26-087 daté du 9 juin 2026