Black-Box Constructions of Protocols for Secure Computation

In this paper, we study the question of whether or not it is possible to construct protocols for general secure computation in the setting of malicious adversaries and no honest majority that use the underlying primitive (e.g., enhanced trapdoor permutation) in a black-box way only. Until now, all known general constructions for this setting were inherently non-black-box since they required the parties to prove zero-knowledge statements that are related to the computation of the underlying primitive. Our main technical result is a fully black-box reduction from oblivious transfer with security against malicious parties to oblivious transfer with security against semihonest parties. As a corollary, we obtain the first constructions of general multiparty protocols (with security against malicious adversaries and without an honest majority) which make only a black-box use of semihonest oblivious transfer, or alternatively a black-box use of lower-level primitives such as enhanced trapdoor permutations or homomorphic encryption. In order to construct this reduction we introduce a new notion of security called privacy in the presence of defensible adversaries. This notion states that if an adversary can produce (retroactively, after the protocol terminates) an input and random tape that make its actions appear to be honest, then it is guaranteed that it learned nothing more than its prescribed output. We then show how to construct defensible oblivious transfer from semihonest oblivious transfer, and malicious oblivious transfer from defensible oblivious transfer, all in a black-box way.