This paper compares the expressive power of five language-based access control models. We show that the expressive powers are incomparable between any pair of history-based access control, regular stack inspection and shallow history automata. Based on these results, we introduce an extension of HBAC, of which expressive power exceeds that of regular stack inspection.