Obfuscating branch decisions based on encrypted data using MISR and hash digests

In this work, we present a novel obfuscation technique that addresses the problem of information leakage in programs with branch decisions based on encrypted data. Our observation is that we can obfuscate individual branch decisions by non-deterministically evaluating all possible execution paths, before “lazily” resolving a digest of the intended branch sequence. We are exploring the effectiveness of signature- and hash-based message digests that are hard to invert and disconnect the branch decisions and the controlling values. Our approach is showcased using a single instruction abstract machine, which is Turing complete and supports branching over encrypted data.