Information Leakage Attacks and Countermeasures

The scientific community has been consistently working on the pervasive problem of information leakage, uncovering numerous attack vectors, and proposing various countermeasures. Despite these efforts, leakage incidents remain prevalent, as the complexity of systems and protocols increases, and sophisticated modeling methods become more accessible to adversaries. This work studies how information leakages manifest in and impact interconnected systems and their users. We first focus on online communications and investigate leakages in the Transport Layer Security protocol (TLS). Using modern machine learning models, we show that an eavesdropping adversary can efficiently exploit meta-information (e.g., packet size) not protected by the TLS’ encryption to launch fingerprinting attacks at an unprecedented scale even under non-optimal conditions. We then turn our attention to ultrasonic communications, and discuss their security shortcomings and how adversaries could exploit them to compromise anonymity network users (even though they aim to offer a greater level of privacy compared to TLS). Following up on these, we delve into physical layer leakages that concern a wide array of (networked) systems such as servers, embedded nodes, Tor relays, and hardware cryptocurrency wallets. We revisit location-based side-channel attacks and develop an exploitation neural network. Our model demonstrates the capabilities of a modern adversary but also presents an inexpensive tool to be used by auditors for detecting such leakages early on during the development cycle. Subsequently, we investigate techniques that further minimize the impact of leakages found in production components. Our proposed system design distributes both the custody of secrets and the cryptographic operation execution across several components, thus making the exploitation of leaks difficult