Redactable graph hashing, revisited

Abstract

We revisit the previous work of Arshad et al. (CODASPY 2014) about the security of redactable graph hashing schemes. Such schemes, introduced in a series of works by Devanbu et al. (DBSec 2000, CCS 2001, Algorithmica 2004), allow to hash graphs and to release sub graphs which can be verified against the original hash value. Arshad et al. introduce security notions for collision resistance and privacy of graphs, where the latter should capture the infeasibility to reconstruct the full graph from the hash value of a redacted one. We discuss here that the original security notions of Arshad et al. are too weak. Our argument is by virtue of intuitively insecure examples which are deemed secure according to their notion. We therefore present stronger security definitions. We also point out the differences in the privacy notions with respect to redactable and sanitizable schemes: In the former case anyone can produce verifiable data from the graph, whereas in the latter case only a designated party can. Sanitizable schemes allow for stronger privacy guarantees. We finally discuss instantiation possibilities for the various security notions.