Security of erasable memories against adaptive adversaries

We study cryptographic modeling and encryption-based design techniques for guaranteeing privacy of data that is first stored in some type of computer memory and then deleted. We continue the investigation started in [3] by presenting an enhanced privacy notion that captures practical scenarios of adversaries repeatedly and adaptively attacking the memory to inspect its entire content before trying to obtain information about deleted data. We prove that the new notion is strictly stronger than the previous one considered in [3] (allowing the adversary a single intrusion), and show then that the efficient protocol in [3] still satisfies the new notion. One question implicitly raised by the previous work was whether it is indeed possible to define one meaningful and applicable notion of security even against adversaries that can repeatedly and adaptively obtain total control of the memory. Perhaps unexpectedly, our paper affirmatively answers this question.