Biblio

Cyber attacks are becoming increasingly frequent, sophisticated, and stealthy. This makes it harder for cyber defence teams to keep up, forcing them to automate their defence capabilities in order to improve their reactivity and efficiency. Therefore, we propose a fully automated cyber defence framework that no longer needs support from humans to detect and mitigate attacks within a complex infrastructure. We design our framework based on a real-world case - Locked Shields - the world's largest cyber defence exercise. In this exercise, teams have to defend their networked infrastructure against attacks, while maintaining operational services for their users. Our framework architecture connects various cyber sensors with network, device, application, and user actuators through an artificial intelligence (AI)-powered automated team in order to dynamically secure the cyber environment. To the best of our knowledge, our framework is the first attempt towards a fully automated cyber defence team that aims at protecting complex environments from sophisticated attacks.

The Dark Web, a conglomerate of services hidden from search engines and regular users, is used by cyber criminals to offer all kinds of illegal services and goods. Multiple Dark Web offerings are highly relevant for the cyber security domain in anticipating and preventing attacks, such as information about zero-day exploits, stolen datasets with login information, or botnets available for hire. In this work, we analyze and discuss the challenges related to information gathering in the Dark Web for cyber security intelligence purposes. To facilitate information collection and the analysis of large amounts of unstructured data, we present BlackWidow, a highly automated modular system that monitors Dark Web services and fuses the collected data in a single analytics framework. BlackWidow relies on a Docker-based micro service architecture which permits the combination of both preexisting and customized machine learning tools. BlackWidow represents all extracted data and the corresponding relationships extracted from posts in a large knowledge graph, which is made available to its security analyst users for search and interactive visual exploration. Using BlackWidow, we conduct a study of seven popular services on the Deep and Dark Web across three different languages with almost 100,000 users. Within less than two days of monitoring time, BlackWidow managed to collect years of relevant information in the areas of cyber security and fraud monitoring. We show that BlackWidow can infer relationships between authors and forums and detect trends for cybersecurity-related topics. Finally, we discuss exemplary case studies surrounding leaked data and preparation for malicious activity.

In recent years, behavioral biometrics have become a popular approach to support continuous authentication systems. Most generally, a continuous authentication system can make two types of errors: false rejects and false accepts. Based on this, the most commonly reported metrics to evaluate systems are the False Reject Rate (FRR) and False Accept Rate (FAR). However, most papers only report the mean of these measures with little attention paid to their distribution. This is problematic as systematic errors allow attackers to perpetually escape detection while random errors are less severe. Using 16 biometric datasets we show that these systematic errors are very common in the wild. We show that some biometrics (such as eye movements) are particularly prone to systematic errors, while others (such as touchscreen inputs) show more even error distributions. Our results also show that the inclusion of some distinctive features lowers average error rates but significantly increases the prevalence of systematic errors. As such, blind optimization of the mean EER (through feature engineering or selection) can sometimes lead to lower security. Following this result we propose the Gini Coefficient (GC) as an additional metric to accurately capture different error distributions. We demonstrate the usefulness of this measure both to compare different systems and to guide researchers during feature selection. In addition to the selection of features and classifiers, some non- functional machine learning methodologies also affect error rates. The most notable examples of this are the selection of training data and the attacker model used to develop the negative class. 13 out of the 25 papers we analyzed either include imposter data in the negative class or randomly sample training data from the entire dataset, with a further 6 not giving any information on the methodology used. Using real-world data we show that both of these decisions lead to significant underestimation of error rates by 63% and 81%, respectively. This is an alarming result, as it suggests that researchers are either unaware of the magnitude of these effects or might even be purposefully attempting to over-optimize their EER without actually improving the system.

In recent years, behavioral biometrics have become a popular approach to support continuous authentication systems. Most generally, a continuous authentication system can make two types of errors: false rejects and false accepts. Based on this, the most commonly reported metrics to evaluate systems are the False Reject Rate (FRR) and False Accept Rate (FAR). However, most papers only report the mean of these measures with little attention paid to their distribution. This is problematic as systematic errors allow attackers to perpetually escape detection while random errors are less severe. Using 16 biometric datasets we show that these systematic errors are very common in the wild. We show that some biometrics (such as eye movements) are particularly prone to systematic errors, while others (such as touchscreen inputs) show more even error distributions. Our results also show that the inclusion of some distinctive features lowers average error rates but significantly increases the prevalence of systematic errors. As such, blind optimization of the mean EER (through feature engineering or selection) can sometimes lead to lower security. Following this result we propose the Gini Coefficient (GC) as an additional metric to accurately capture different error distributions. We demonstrate the usefulness of this measure both to compare different systems and to guide researchers during feature selection. In addition to the selection of features and classifiers, some non- functional machine learning methodologies also affect error rates. The most notable examples of this are the selection of training data and the attacker model used to develop the negative class. 13 out of the 25 papers we analyzed either include imposter data in the negative class or randomly sample training data from the entire dataset, with a further 6 not giving any information on the methodology used. Using real-world data we show that both of these decisions lead to significant underestimation of error rates by 63% and 81%, respectively. This is an alarming result, as it suggests that researchers are either unaware of the magnitude of these effects or might even be purposefully attempting to over-optimize their EER without actually improving the system.

Multilateration techniques have been proposed to verify the integrity of unprotected location claims in wireless localization systems. A common assumption is that the adversary is equipped with only a single device from which it transmits location spoofing signals. In this paper, we consider a more advanced model where the attacker is equipped with multiple devices and performs a geographically distributed coordinated attack on the multilateration system. The feasibility of a distributed multi-device attack is demonstrated experimentally with a self-developed attack implementation based on multiple COTS software-defined radio (SDR) devices. We launch an attack against the OpenSky Network, an air traffic surveillance system that implements a time-difference-of-arrival (TDoA) multi-lateration method for aircraft localization based on ADS-B signals. Our experiments show that the timing errors for distributed spoofed signals are indistinguishable from the multilateration errors of legitimate aircraft signals, indicating that the threat of multi-device spoofing attacks is real in this and other similar systems. In the second part of this work, we investigate physical-layer features that could be used to detect multi-device attacks. We show that the frequency offset and transient phase noise of the attacker's radio devices can be exploited to discriminate between a received signal that has been transmitted by a single (legitimate) transponder or by multiple (malicious) spoofing sources. Based on that, we devise a multi-device spoofing detection system that achieves zero false positives and a false negative rate below 1%.

Future transportation systems highly rely on the integrity of spatial information provided by their means of transportation such as vehicles and planes. In critical applications (e.g. collision avoidance), tampering with this data can result in life-threatening situations. It is therefore essential for the safety of these systems to securely verify this information. While there is a considerable body of work on the secure verification of locations, movement of nodes has only received little attention in the literature. This paper proposes a new method to securely verify spatial movement of a mobile sender in all dimensions, i.e., position, speed, and direction. Our scheme uses Doppler shift measurements from different locations to verify a prover's motion. We provide formal proof for the security of the scheme and demonstrate its applicability to air traffic communications. Our results indicate that it is possible to reliably verify the motion of aircraft in currently operational systems with an equal error rate of zero.