Cryptographic Protocols Provably Secure Against Dynamic Adversaries

We introduce new techniques for generating and reasoning about protocols. These techniques are based on protocol transformations that depend on the nature of the adversaries under consideration. We propose a set of definitions that, captures and unifies the intuitive notions of correctness, privacy, and robustness, and enables us to give concise and modular proofs that our protocols possess these desirable properties. Using these techniques, whose major purpose is to greatly simplify the design and verification of cryptographic protocols, we show how to construct a multiparty cryptographic protocol to compute any given feasible function of the parties' inputs. We prove that our protocol is secure against the malicious actions of any adversary, limited to feasible computation, but with the power to eavesdrop on all messages and to corrupt any dynamically chosen minority of the parties. This is the first proof of security against dynamic adversaries in the "cryptographic" model of multiparty protocols. We assume the existeuce of a one-way function and allow the participants to erase small portions of memory. Our result combines the superior resilience of the cryptographic setting of [GMW87] with the stronger (dynamic) fault pattern of the "non-cryptographic" setting of [BGW88, CCD88].