Privacy-Preserving Implicit Authentication

Abstract

Abstract. In an implicit authentication system, a user profile is used as an additional factor to strengthen the authentication of mobile users. The profile consists of features that are constructed using the history of user actions on her mobile device over time. The profile is stored on the server and is used to authenticate an access request originated from the device at a later time. An access request will include a vector of recent measurements of the features on the device, that will be subsequently matched against the features stored at the server, to accept or reject the request. The features however include private information such as user location or web sites that have been visited. We propose a privacy-preserving implicit authentication system that achieves implicit authentication without revealing information about the usage profiles of the users to the server. We propose an architecture, give a formal security model and a construction with provable security in two settings where: (i) the device follows the protocol, and (ii) the device is captured and behaves maliciously.

Flexible and Robust Privacy-Preserving Implicit Authentication

“…The protocol in [17] needs the server to store the users' accumulated usage profiles encrypted under two different cryptosystems, one that is homomorphic and one that is order-preserving. We present here a protocol for privacy-preserving implicit authentication based on set intersection, which has the advantage that the server only needs to store the users' accumulated usage profiles encrypted under one (homomorphic) cryptosystem.…”

“…This allows saving storage at the carrier and also computation during the protocol. Also, unlike [17], our protocol does not leak the relative order of fresh feature samples collected by the user's device for comparison with the encrypted profile. Finally, our protocol can deal with any type of features (numerical or non-numerical), while the protocol [17] is restricted to numerical features.…”

“…The rest of this paper is organized as follows. Section 2 gives background on implicit authentication and on the privacy-preserving implicit authentication protocol of [17]. Section 3 discusses how to compute the dissimilarity between two sets depending on the type of their elements.…”

“…In the privacy-preserving implicit authentication system proposed in [17], the user's device encrypts the user's usage profile at set-up time, and forwards it to the carrier, who stores it for later comparison. There is no security problem because during normal operation the user's device does not store the user's profile (it just collects the fresh usage features).…”

“…The core of proposal [17] is the algorithm for computing the dissimilarity score between two inputs: the fresh sample provided by the user's device and the profile stored at the carrier. All the computation takes place at the carrier and both inputs are encrypted: indeed, the carrier stores the encrypted profile and the user's device sends the encrypted fresh sample to the carrier.…”