Screaming-channel attacks enable Electromagnetic (EM) Side-Channel Attacks (SCAs) at larger distances due to higher EM leakage energies than traditional SCAs, relaxing the requirement of close access to the victim. This attack can be mounted on devices integrating Radio Frequency (RF) modules on the same die as digital circuits, where the RF can unintentionally capture, modulate, amplify, and transmit the leakage along with legitimate signals. Leakage results from digital switching activity, so the hypothesis of previous works was that this leakage would appear at multiples of the digital clock frequency, i.e., harmonics. Our work 14 demonstrates that compromising signals appear not only at the harmonics and that leakage at non-harmonics can be exploited for successful attacks. Indeed, the transformations undergone by the leaked signal are complex due to propagation effects through the substrate and power and ground planes, so the leakage also appears at other frequencies. We first propose two methodologies to locate frequencies that contain leakage and demonstrate that it appears at non-harmonic frequencies. Then, our experimental results show that screaming-channel attacks at non-harmonic frequencies can be as successful as at harmonics when retrieving a 16-byte AES key. As the RF spectrum is polluted by interfering signals, we run experiments and show successful attacks in a more realistic, noisy environment where harmonic frequencies are contaminated by multi-path fading and interference. These attacks at non-harmonic frequencies increase the attack surface by providing attackers with an increased number of potential frequencies where attacks can succeed.

On-board payload data processing can be performed by developing space-qualified heterogeneous Multiprocessor Systemon-Chips (MPSoCs). We present in 19 key compute-intensive payload algorithms, based on a survey with space science researchers, including the two-dimensional Fast Fourier Transform (2-D FFT). Also, we propose to perform design space exploration by combining the roofline performance model with High-Level Synthesis (HLS) for hardware accelerator architecture design. The roofline model visualizes the limits of a given architecture regarding Input/Output (I/O) bandwidth and computational performance, along with the achieved performance for different implementations. HLS is an interesting option in developing FPGA-based onboard processing applications for payload teams that need to adjust architecture specifications through design reviews and have limited expertise in Hardware Description Languages (HDLs). In this paper, we focus on an FPGA-based MPSoC thanks to recently released radiation-hardened heterogeneous embedded platforms.

Analyzing Android applications is essential to review proprietary code and to understand malware behaviors. However, Android applications use obfuscation techniques to slow down this process. These obfuscation techniques are increasingly based on native code. In 6, we propose OATs'inside, a new analysis tool that focuses on high-level behaviors to circumvent native obfuscation techniques transparently. The targeted high-level behaviors are object-level behaviors, i.e., actions performed on Java objects (e.g., field accesses, method calls), regardless of whether these actions are performed using Java or native code. Our system uses a hybrid approach based on dynamic monitoring and trace-based symbolic execution to output control flow graphs (CFGs) for each method of the analyzed application. CFGs are composed of Java-like actions enriched with condition expressions and dataflows between actions, giving an understandable representation of any code, even those fully native. OATs'inside spares users the need to dive into low-level instructions, which are difficult to reverse engineer. We extensively compare OATs'inside functionalities against state-of-the-art tools to highlight the benefit when observing native operations. Our experiments are conducted on a real smartphone: We discuss the performance impact of OATs'inside, and we demonstrate its practical use on applications containing anti-debugging techniques provided by the OWASP foundation. We also evaluate the robustness of OATs'inside using obfuscated unit tests using the Tigress obfuscator.

Malware analysis consists of studying a sample of suspicious code to understand it and producing a representation or explanation of this code that can be used by a human expert or a clustering/classification/detection tool. The analysis can be static (only the code is studied) or dynamic (only the interaction between the code and its host during one or more executions is studied). The quality of the interpretation of a code and its later detection depends on the quality of the information contained in this representation. To date, many analyses produce voluminous reports that are difficult to handle quickly. In 23, we present BAGUETTE, a graph-based representation of the interactions of a sample and the resources offered by the host system during one execution. We explain how BAGUETTE helps automatically search for specific behaviors in a malware database and how it efficiently assists the expert in analyzing samples.

Today, the classification of a file as either benign or malicious is performed by a combination of deterministic indicators (such as antivirus rules), machine learning classifiers, and, more importantly, the judgment of human experts. However, to compare the difference between human and machine intelligence in malware analysis, it is first necessary to understand how human subjects approach malware classification. In this direction, we present in 7 the first experimental study designed to capture which ‘features’ of a suspicious program (e.g., static properties or runtime behaviors) are prioritized for malware classification according to humans and machines intelligence. For this purpose, we created a malware classification game where 110 human players worldwide and with different seniority levels (72 novices and 38 experts) have competed to classify the highest number of unknown samples based on detailed sandbox reports. Surprisingly, we discovered that both experts and novices base their decisions on approximately the same features, even if there are clear differences between the two expertise classes. Furthermore, we implemented two state-of-the-art Machine Learning models for malware classification and evaluated their performances on the same set of samples. The comparative analysis of the results unveiled a common set of features preferred by both Machine Learning models and helped better understand the difference in the feature extraction. This work reflects the difference in the decision-making process of humans and computer algorithms and the different ways they extract information from the same data. Its findings serve multiple purposes, from training better malware analysts to improving feature encoding.

Many studies have proposed machine-learning (ML) models for malware detection and classification, reporting an almost-perfect performance. However, they assemble ground-truth in different ways, use diverse static-and dynamic-analysis techniques for feature extraction, and even differ on what they consider a malware family. As a consequence, our community still lacks an understanding of malware classification results: whether they are tied to the nature and distribution of the collected dataset, to what extent the number of families and samples in the training dataset influence performance, and how well static and dynamic features complement each other. The article 12 sheds light on those open questions by investigating the impact of datasets, features, and classifiers on ML-based malware detection and classification. For this, we collect the largest balanced malware dataset so far with 67k samples from 670 families (100 samples each), and train state-of-the-art models for malware detection and family classification using our dataset. Our results reveal that static features perform better than dynamic features, and that combining both only provides marginal improvement over static features. We discover no correlation between packing and classification accuracy, and that missing behaviors in dynamically-extracted features highly penalise their performance. We also demonstrate how a larger number of families to classify makes the classification harder, while a higher number of samples per family increases accuracy. Finally, we find that models trained on a uniform distribution of samples per family better generalize on unseen data.

Federated learning (FL) enables multiple parties to collaboratively train a machine learning model without sharing their data; rather, they train their own model locally and send updates to a central server for aggregation. Depending on how the data is distributed among the participants, FL can be classified into Horizontal (HFL) and Vertical (VFL). In VFL, the participants share the same set of training instances but only host a different and non-overlapping subset of the whole feature space. Whereas in HFL, each participant shares the same set of features while the training set is split into locally owned training data subsets. VFL is increasingly used in applications like financial fraud detection; nonetheless, very little work has analyzed its security. In 20, we focus on robustness in VFL, in particular, on backdoor attacks, whereby an adversary attempts to manipulate the aggregate model during the training process to trigger misclassifications. Performing backdoor attacks in VFL is more challenging than in HFL, as the adversary i) does not have access to the labels during training and ii) cannot change the labels as she only has access to the feature embeddings. We present a first-of-its-kind clean-label backdoor attack in VFL, which consists of two phases: a label inference and a backdoor phase. We demonstrate the effectiveness of the attack on three different datasets, investigate the factors involved in its success, and discuss countermeasures to mitigate its impact.

When an attacker targets a system, he aims to remain undetected as long as possible. He must therefore avoid performing actions that are characteristic of an identified malicious behavior. One way to avoid detection is to only perform actions on the system that appear legitimate. That is, actions that are allowed because of the system configuration or actions that are possible by diverting the use of legitimate services. In 21, we present and experiment with AWARE (Attacks in Windows Architectures REvealed), a defensive tool able to query a Windows system and build a directed graph highlighting possible stealthy attack paths that an attacker could use during the propagation phase of an attack campaign. These attack paths only rely on legitimate system actions and the use of Living-Off-The-Land binaries. AWARE also proposes a range of corrective measures to prevent these attack paths.

In cybersecurity, CVEs (Common Vulnerabilities and Exposures) are publicly disclosed hardware or software vulnerabilities. These vulnerabilities are documented and listed in the NVD database maintained by the NIST. Knowledge of the CVEs impacting an information system provides a measure of its level of security. In 22 we point out that these vulnerabilities should be described in greater detail to understand how they could be chained together in a complete attack scenario. This article presents the first proposal for the CAPG format, which is a method for representing a CVE vulnerability, a corresponding exploit, and associated attack positions.

The use of machine learning for anomaly detection in cyber security-critical applications, such as intrusion detection systems, has been hindered by the lack of explainability. Without understanding the reason behind anomaly alerts, it is too expensive or impossible for human analysts to verify and identify cyber-attacks. Our research addresses this challenge and focuses on unsupervised network intrusion detection, where only benign network traffic is available for training the detection model. In 18, we propose a novel post-hoc explanation method, called AE-pvalues, which is based on the p-values of the reconstruction errors produced by an Auto-Encoder-based anomaly detection method. Our work identifies the most informative network traffic features associated with an anomaly alert, providing interpretations for the generated alerts. We conduct an empirical study using a large-scale network intrusion dataset, CICIDS2017, to compare the proposed AE-pvalues method with two state-of-the-art baselines applied in the unsupervised anomaly detection task. Our experimental results show that the AE-pvalues method accurately identifies abnormal influential network traffic features. Furthermore, our study demonstrates that the explanation outputs can help identify different types of network attacks in the detected anomalies, enabling human security analysts to understand the root cause of the anomalies and take prompt action to strengthen security measures.

In recent years, the disclosure of several significant security vulnerabilities has revealed the trust put in some presumed security properties of commonplace hardware to be misplaced. We propose to design hardware systems with security mechanisms, together with a formal statement of the security properties obtained, and a machine-checked proof that the hardware security mechanisms indeed implement the sought-for security property. Formally proving security properties about hardware systems might seem prohibitively complex and expensive. In 8, we tackle this concern by designing a realistic and accessible methodology on top of the Kôika Hardware Description Language 27 for specifying and proving security properties during hardware development. Our methodology is centered around a verified compiler from high-level and inefficient to work with Kôika models to an equivalent lower-level representation where side effects are made explicit and reasoning is convenient. We apply this methodology to a concrete example: the formal specification and implementation of a shadow stack mechanism on an RV32I processor. We prove that this security mechanism is correct, i.e., any illegal modification of a return address does indeed result in the termination of the whole system. Furthermore, we show that this modification of the processor does not impact its behaviour in other, unexpected ways.

The recent rise of interest in distributed applications has highlighted the importance of effective information dissemination. The challenge lies in the fact that nodes in a distributed system are not necessarily synchronized, and may fail at any time. This has led to the emergence of randomized rumor spreading protocols, such as push and pull protocols, which have been studied extensively. The $k$-pull operation, which allows an uninformed node to ask for the rumor from a fixed number of other nodes in parallel, has been proposed to improve the pull algorithm's effectiveness. In 16, we present and study the performance of the $k$-pull operation in the presence of a certain fraction $f$ of non-cooperative nodes. Our goal is to understand the impact of $k$ on the propagation of the rumor despite the presence of a fraction $f$ of non-collaborative nodes.

Contrary to intuition, insecure computer network architectures are valuable assets in IT security. Indeed, such architectures (referred to as cyber-ranges) are commonly used to train red teams and test security solutions, in particular the ones related to supervision security. Unfortunately, the design and deployment of these cyber-ranges is costly, as they require designing an attack scenario from scratch and then implementing it in an architecture on a case-by-case basis, through manual choices of machines/users, OS versions, available services and configuration choices. The article 10 presents URSID, a framework for automatic deployment of cyber-ranges based on the formal description of attack scenarios. The scenario is described at the technical attack level according to the MITRE nomenclature, refined into several variations (instances) at the procedural level and then deployed in virtual multiple architectures. URSID thus automates costly manual tasks and allows to have several instances of the same scenario on architectures with different OS, software or account configurations. URSID has been successfully tested in an academic cyber attack and defense training exercise as detailed in Section 10.3.2.

The long-term feasibility of blockchain technology is hindered by the inability of existing blockchain protocols to prune the consensus data leading to constantly growing storage and communication requirements. Kiayias et al. have proposed Non-Interactive-Proofs-of-Proof-of-Works (NIPoPoWs) as a mechanism to reduce the storage and communication complexity of blockchains to O(poly log(n)). However, their protocol is only resilient to an adversary that may control strictly less than a third of the total computational power, which is a reduction from the security guaranteed by Bitcoin and other existing Proof-ofbased blockchains. In 15, we present an improvement to the Kiayias et al. proposal, which is resilient against an adversary that may control less than half of the total computational power while operating in O(poly log(n)) storage and communication complexity. Additionally, we present a novel proof that establishes a lower bound of O(log(n)) on the storage and communication complexity of any PoW-based blockchain protocol.