AI agents can act autonomously, but current laws generally hold humans or businesses responsible for harm, raising new questions about AI liability.