AI developers should be legally responsible for unauthorised actions performed by their autonomous agents