Key regression from constrained pseudorandom functions

Abstract

Key regression is a primitive that allows a content publisher to give a small amount of keying material, for time stamp i, to a party such that they can derive key ki and all keys kj for time stamps j<i, but not for time stamps j>i. Existing key-regression schemes based on random oracles, pseudorandom permutations, and pseudorandom generators require a publisher to maintain state of size linear in the number of potential keys. In previous work, constant sized publisher state has only been accessible through the use of trapdoor permutations (e.g., RSA). This paper presents a novel key-regression scheme where the publisher must store only a constant amount of state information without trapdoor permutations. This is achieved through the use of constrained pseudorandom functions, a primitive that allows for the delegation of partial evaluation without revealing the key.