ABSTRACT
As AI systems evolve from assistive tools into increasingly autonomous, agentic actors, long-standing doctrines of corporate liability are being put to the test. This Essay examines whether existing legal frameworks can hold corporations accountable for harms caused by advanced forms of AI. It suggests that, first, existing private law doctrines continue to provide a largely workable foundation without recognizing AI agency or personhood, by focusing on design, deployment, supervision, and organizational duties. Second, a doctrinal obstacle arises in systems of corporate attribution that depend on discrete human wrongs. This model becomes more fragile with agentic AI, confirming the need for depersonalized forms of corporate liability grounded in direct entity-level duties, collective wrongdoing, and organizational failures to ensure public safety. Third, even where attribution remains possible, accountability gaps persist, particularly regarding evidentiary asymmetries stemming from algorithmic opacity, difficulties in establishing causation and foreseeability, multi-actor AI ecosystems, and jurisdictional fragmentation. The Essay concludes that existing doctrines are broadly adaptable and gaps likely to be limited. Nevertheless, selective legislative measures are desirable to complement and adapt as necessary existing private law tools.
Petrin, Martin, When AI Hurts: Corporate Responsibility for Agentic AI (May 1, 2026), Harvard Journal of Law & Technology, volume 40 (forthcoming).
Leave a Reply