Idealizing Identity-Based Encryption

We formalize the standard application of identity-based encryption (IBE), namely noninteractive secure communication, as realizing an ideal system which we call delivery controlled channel (DCC). This system allows users to be registered (by a central authority) for an identity and to send messages securely to other users only known by their identity. Quite surprisingly, we show that existing security definitions for IBE are not sufficient to realize DCC. In fact, it is impossible to do so in the standard model. We show, however, how to adjust any IBE scheme that satisfies the standard security definition IND-ID-CPA to achieve this goal in the random oracle model. We also show that the impossibility result can be avoided in the standard model by considering a weaker ideal system that requires all users to be registered in an initial phase before any messages are sent. To achieve this, a weaker security notion, which we introduce and call IND-ID1-CPA, is actually sufficient. This justifies our new security definition and might open the door for more efficient schemes. We further investigate which ideal systems can be realized with schemes satisfying the standard notion and variants of selective security. As a contribution of independent interest, we show how to model features of an ideal system that are potentially available to dishonest parties but not guaranteed, and which such features arise when using IBE.