The abundance of data and ease of availability of third-party storage solutions, has led people to store their private and sensitive data on the cloud. These cloud services also allow us to do searches and updates over our data without having to download all of it to our devices. Hence, these services not only see our data in the clear but also all our searches and updates. Even if we choose to trust the cloud, data breaches are an ever-present threat. Over the last twenty years, many government institutions as well as private companies, big and small, have been targeted for people’s data including but not limited to financial and identity-related information, medical records, etc. Motivated by this privacy loss and the constant occurrence of data breaches, we study the problem of encrypted search, where a client outsources data to an untrusted server while maintaining the ability to privately query and dynamically update the stored data. From a privacy perspective, the goal is to protect the data and the operations on it from the server as well as from other unwanted parties such as hackers who could breach the server. We model real world threats, provide new formal security definitions and design new practical solutions with reasonable leakage profiles that are provably-secure under our security definitions. Specifically, we deal with persistent, snapshot and tenacious adversaries, to name a few, that capture different real-world adversarial threats. We design schemes that are secure against these different adversaries. The constructions we propose guarantee several standard security properties for dynamic searchable encryption schemes including volume-hiding, forward and backward privacy. We also introduce formal definitions for security properties such as dynamic volume-hiding and for novel properties such as past-hiding and correlation security that thwart new kinds of attacks that have not been considered in the past. We also implemented our constructions and evaluated their concrete efficiency empirically. Our experiments show that our constructions are highly efficient, where the search protocol in some schemes takes less than 1 microsecond per query/result pair.
