The rise of model sharing through frameworks and dedicated hubs makes Machine Learning significantly more accessible. Despite its benefits, loading shared models exposes users to underexplored security risks, while security awareness remains limited among both practitioners and developers. To enable a more security-conscious approach in Machine Learning model sharing, in this paper, we evaluate the security posture of frameworks and hubs, assess whether security-oriented mechanisms offer real protection, and survey how users perceive the security narratives surrounding model sharing. Our evaluation shows that most frameworks and hubs address security risks partially at best, often by shifting responsibility to the user. More concerningly, our analysis of frameworks advertising security-oriented settings and complete model sharing uncovered multiple 0-day vulnerabilities enabling arbitrary code execution. Through this analysis, we show that, despite the recent narrative, securely loading Machine Learning models is far from being a solved problem and cannot be guaranteed by the file format used for sharing. Our survey shows that the security narrative leads users to consider security-oriented settings as trustworthy, despite the weaknesses shown in this work. From this, we derive suggestions to strengthen the security of model-sharing ecosystems.

Federated Learning has emerged as a privacy-oriented alternative to centralized Machine Learning, enabling collaborative model training without direct data sharing. While extensively studied for neural networks, the security and privacy implications of tree-based models remain underexplored. This work introduces TimberStrike, an optimization-based dataset reconstruction attack targeting horizontally federated tree-based models. Our attack, carried out by a single client, exploits the discrete nature of decision trees by using split values and decision paths to infer sensitive training data from other clients. We evaluate TimberStrike on State-of-the-Art federated gradient boosting implementations across multiple frameworks, including Flower, NVFlare, and FedTree, demonstrating their vulnerability to privacy breaches. On a publicly available stroke prediction dataset, TimberStrike consistently reconstructs between 73.05% and 95.63% of the target dataset across all implementations. We further analyze Differential Privacy, showing that while it partially mitigates the attack, it also significantly degrades model performance. Our findings highlight the need for privacy-preserving mechanisms specifically designed for tree-based Federated Learning systems, and we provide preliminary insights into their design.

The security of modern vehicles has become increasingly important, with the controller area network (CAN) bus serving as a critical communication backbone for various electronic control units (ECUs). The absence of robust security measures in CAN, coupled with the increasing connectivity of vehicles, makes them susceptible to cyberattacks. While intrusion detection systems (IDSs) have been developed to counter such threats, they are not foolproof. Adversarial attacks, particularly evasion attacks, can manipulate inputs to bypass detection by IDSs. This article extends our previous work by investigating the feasibility and impact of gradient-based adversarial attacks performed with different degrees of knowledge against automotive IDSs. We consider three scenarios: white-box (attacker with full system knowledge), grey-box (partial system knowledge), and—the more realistic—black-box (no knowledge of the IDS’s internal workings or data). We evaluate the effectiveness of the proposed attacks against state-of-the-art IDSs on two publicly available datasets. Additionally, we study the effect of the adversarial perturbation on the attack impact and evaluate real-time feasibility by precomputing evasive payloads for timed injection based on bus traffic. Our results demonstrate that, besides attacks being challenging due to the automotive domain constraints, their effectiveness is strongly dependent on the dataset quality, the target IDS, and the attacker’s degree of knowledge.

The complexities introduced by compiler optimization have long stood as a significant obstacle in binary analysis and reverse engineering. Function inlining, in particular, complicates function recognition by replacing function calls with the entire body of the callee, mixing code from multiple functions. State-of-the-art approaches can identify inlined functions at basic block granularity, but cannot determine which instructions belong to each function and precisely deduce inlined boundaries. Without this information, further analyses such as decompilation cannot be performed effectively. This paper presents Highliner, a novel approach that improves state-of-the-art approaches by identifying inline instances at instruction-level granularity. Highliner operates downstream of block-level detectors: given basic blocks reported by state-of-the-art approaches as belonging to a specific inlined function, it labels each instruction as Inlined or Not inlined and recovers the inlined-function boundaries. We treat the problem as a sequence tagging task typical of NLP and implement a learning-based technique involving instruction embedding and recurrent neural networks. We compile a dataset of open-source projects with different optimizations and use the DWARF debug information standard to construct labeled sequences of inline instructions. We use this dataset to train, validate, and test a sequence labeling architecture in which instructions are encoded via the pre-trained assembly language transformer PalmTree and then processed by an RNN-based classifier to produce binary predictions. When evaluated as a binary classifier, Highliner achieves an F1-score of 0.94 overall. In addition, when specifically tested on recognizing function boundaries, Highliner achieves an Accuracy of 0.82 on initial boundaries and 0.83 on final boundaries.

In aviation, safety is paramount, with air traffic control (ATC) playing a crucial role in monitoring aircraft to prevent collisions and manage traffic flows. In response to increasing air traffic, a renewal process has been initiated. This includes deploying the automatic dependent surveillance-broadcast (ADS-B) communications protocol, which aims to enhance surveillance precision and increase the number of aircraft that can be handled simultaneously. This transition is transforming ATC from a radar-based system to a more advanced satellite-based global positioning system (GPS) location tracking system. However, due to its inherently open design, the ADS-B protocol lacks critical security features such as authentication, necessitating the adoption of additional security measures to mitigate potential cyber-attacks. To address these vulnerabilities, this work introduces Swarm, an innovative distributed ledger-based framework, built on top of the ADS-B protocol and aimed at enhancing the security of air traffic control (ATC) while avoiding single points of failure. Swarm can be integrated into existing ATC infrastructure without requiring any modifications to the ADS-B protocol. We evaluate Swarm through rigorous and realistic attack scenarios, using real-world aviation data, demonstrating its capability to enhance the security of the aviation domain.

Address Space Layout Randomization (ASLR) is a crucial defense mechanism employed by modern operating systems to mitigate exploitation by randomizing processes? memory layouts. However, the stark reality is that real-world implementations of ASLR are imperfect and subject to weaknesses that attackers can exploit. This work evaluates the effectiveness of ASLR on major desktop platforms, including Linux, MacOS, and Windows, by examining the variability in the placement of memory objects across various processes, threads, and system restarts. In particular, we collect samples of memory object locations, conduct statistical analyses to measure the randomness of these placements and examine the memory layout to find any patterns among objects that could decrease this randomness. The results show that while some systems, like Linux distributions, provide robust randomization, others, like Windows and MacOS, often fail to adequately randomize key areas like executable code and libraries. Moreover, we find a significant entropy reduction in the entropy of libraries after the Linux 5.18 version and identify correlation paths that an attacker could leverage to reduce exploitation complexity significantly. Ultimately, we rank the identified weaknesses based on severity and validate our entropy estimates with a proof-of-concept attack. In brief, this paper provides the first comprehensive evaluation of ASLR effectiveness across different operating systems and highlights opportunities for Operating System (OS) vendors to strengthen ASLR implementations.

The rapid evolution of artificial intelligence (AI) technologies holds transformative potential for the healthcare sector. In critical situations requiring immediate decision-making, healthcare professionals can leverage machine learning (ML) algorithms to prioritize and optimize treatment options, thereby reducing costs and improving patient outcomes. However, the sensitive nature of healthcare data presents significant challenges in terms of privacy and data ownership, hindering data availability and the development of robust algorithms. Federated Learning (FL) addresses these challenges by enabling collaborative training of ML models without the exchange of local data. This paper introduces a novel FL platform designed to support the configuration, monitoring, and management of FL processes. This platform operates on Platform-as-a-Service (PaaS) principles and utilizes the Message Queuing Telemetry Transport (MQTT) publish-subscribe protocol. Considering the production readiness and data sensitivity inherent in clinical environments, we emphasize the security of the proposed FL architecture, addressing potential threats and proposing mitigation strategies to enhance the platform's trustworthiness. The platform has been successfully tested in various operational environments using a publicly available dataset, highlighting its benefits and confirming its efficacy.

This paper analyzes 88 million hacker forum posts of a publicly available dataset and 75,000 online articles over a 20-year timespan, studying the potential of hacker forums as a proactive Cyber Threat Intelligence (CTI) source. Using a custom Natural Language Processing pipeline with fine-tuned BERT-based models, we extract named entities from forum posts and reports and cross-reference their date of occurrence over different periods. Our analysis reveals that discussions on hacker forums precede official security reports for over 60% of the identified entities in 20 years of data. This highlights the relevance of these platforms as early indicators of cyber threats. However, our longitudinal analysis shows that such a trend has been constantly decreasing since 2012: forum discussions no longer consistently anticipate threats discussed in cybersecurity reports, possibly due to increased scrutiny or the emergence of alternative channels. This suggests that the CTI community should adapt by identifying and monitoring new platforms where threat actors congregate. Despite not being as thriving as in the first decade of 2000, underground communities are still releasing novel malware and showing interest in discussing malware employed in real cyberattacks. Our results highlight the value of hacker forums as early threat indicators and the importance of proactively monitoring them for potential cyberattack detection. This approach addresses the research gap that predominantly focuses on traditional cybersecurity reports.

Several critical contexts, such as healthcare, smart cities, drones, transportation, and agriculture, nowadays rely on IoT, or more in general embedded, devices that require comprehensive security analysis to ensure their integrity before deployment. Security concerns are often related to vulnerabilities that result from inadequate coding or undocumented features that may create significant privacy issues for users and companies. Current analysis methods, albeit dependent on complex tools, may lead to superficial assessments due to compatibility issues, while authoritative entities struggle with specifying feasible firmware analysis requests for manufacturers within operational contexts. This paper urges the scientific community to collaborate with stakeholders—manufacturers, vendors, security analysts, and experts—to forge a cooperative model that clarifies manufacturer contributions and aligns analysis demands with operational constraints. Aiming at a modular approach, this paper highlights the crucial need to refine security analysis, ensuring more precise requirements, balanced expectations, and stronger partnerships between vendors and analysts. To achieve this, we propose a threat model based on the feasible interactions of actors involved in the security evaluation of a device, with a particular emphasis on the responsibilities and necessities of all entities involved.

The challenges derived from the data-intensive nature of machine learning in conjunction with technologies that enable novel paradigms such as V2X and the potential offered by 5G communication, allow and justify the deployment of Federated Learning (FL) solutions in the vehicular intrusion detection domain. In this paper, we investigate the effects of integrating FL strategies into the machine learning-based intrusion detection process for on-board vehicular networks. Accordingly, we propose a FL implementation of a state-of-the-art Intrusion Detection System (IDS) for Controller Area Network (CAN), based on LSTM autoencoders. We thoroughly evaluate its detection efficiency and communication overhead, comparing it to a centralized version of the same algorithm, thereby presenting it as a feasible solution.

Federated Learning (FL) has emerged as a promising approach for privacy-preserving machine learning, particu-larly in sensitive domains such as healthcare. In this context, the TRUSTroke project aims to leverage FL to assist clinicians in ischemic stroke prediction. This paper provides an overview of the TRUSTroke FL network infrastructure. The proposed archi-tecture adopts a client-server model with a central Parameter Server (PS). We introduce a Docker-based design for the client nodes, offering a flexible solution for implementing FL processes in clinical settings. The impact of different communication pro-tocols (HTTP or MQTT) on FL network operation is analyzed, with MQTT selected for its suitability in FL scenarios. A control plane to support the main operations required by FL processes is also proposed. The paper concludes with an analysis of security aspects of the FL architecture, addressing potential threats and to increase trustworthiness.

Fuzzing is a dynamic analysis technique that repeatedly executes the target program with many different inputs to trigger abnormal behavior, such as a crash. One of the most successful techniques consists in generating inputs to increase code-coverage by using a mutational approach: this type of fuzzers maintains a population of inputs, they perform mutations on the inputs in the current population, and they add mutated inputs to the population if they discover new code-coverage in the target program. Researchers are continuously looking for techniques to increment the efficiency of fuzzers; one of these techniques consists in generating heat-maps for targeting specific bytes during the mutation of the input, as not all bytes might be useful for controlling the program's workflow. We propose the first approach in the literature that uses reinforcement learning for building heat-maps, by formalizing the problem of choosing the position to be mutated within the input as a reinforcement-learning problem. We model the policy by means of a neural network, and we train it by using Proximal Policy Optimization (PPO). We implement our approach in Rainfuzz, and we show the effectiveness of its heat-maps by comparing Rainfuzz against an equivalent fuzzer that performs mutations at random positions. We achieve the best performance by running AFL++ and Rainfuzz in parallel (in a collaborative fuzzing setting), outperforming a setting where we run two AFL++ instances in parallel.

In this paper, we propose BINO, a static analysis approach that relieves reverse engineers from the challenging task of recognizing library functions that have been inlined. BINOrecognizes inline calls of methods of C++ template classes (even with unknown data types). We do this through a binary fingerprinting and matching approach. Our fingerprint model captures syntactic and semantic features of an assembly function, along with its Control-Flow Graph structure. Using these fingerprints and subgraph isomorphism, it recognizes inline method calls in a target binary. BINOautomates the fingerprints generation phase by parsing the source code of the template classes and automatically building appropriate binaries with representative inline calls of said methods. We evaluate BINOby performing experiments on a dataset of 555 GitHub C++ projects containing 10600 inline functions, exploring several optimization levels that allow the compiler to inline function calls. We show that our approach can recognize inline function calls to the most used methods of well-known template classes with an F1-Score up to 63% with the -O2, -O3, and -Ofast optimizations levels.

Over the years, vehicles have become increasingly complex and an attractive target for malicious adversaries. This raised the need for effective and efficient Intrusion Detection Systemss (IDSs) for onboard networks able to work with the stringent requirements and the heterogeneity of information transmitted on the Controller Area Network. While state-of-the-art solutions are effective in detecting specific types of anomalies and work on a subset of the CAN signals, no single method can perform better than the others on all types of attacks, particularly if they need to provide predictions to comply with the domain’s real-time constraints. In this paper, we present CANova, a modular framework that exploits the characteristics of the different Controller Area Network (CAN) packets to select the Intrusion Detection Systemss (IDSs) that better fits them. In particular, it uses flow- and payload-based IDSs to analyze the packets’ content and arrival time. We evaluate CANova by comparing its performance against state-of-the-art Intrusion Detection Systemss (IDSs) for in-vehicle network and a comprehensive set of synthetic and real attacks in real-world CAN datasets. We demonstrate that our approach can achieve good performances in terms of detection, false positive rates, and temporal performances.

Machine learning (ML) models are vulnerable to adversarial machine learning (AML) attacks. Unlike other contexts, the fraud detection domain is characterized by inherent challenges that make conventional approaches hardly applicable. In this article, we extend the application of AML techniques to the fraud detection task by studying poisoning attacks and their possible countermeasures. First, we present a novel approach for performing poisoning attacks that overcomes the fraud detection domain-specific constraints. It generates fraudulent candidate transactions and tests them against a machine learning-based Oracle , which simulates the target fraud detection system aiming at evading it. Misclassified fraudulent candidate transactions are then integrated into the target detection system’s training set, poisoning its model and shifting its decision boundary. Second, we propose a novel approach that extends the adversarial training technique to mitigate AML attacks: During the training phase of the detection system, we generate artificial frauds by modifying random original legitimate transactions; then, we include them in the training set with the correct label. By doing so, we instruct our model to recognize evasive transactions before an attack occurs. Using two real bank datasets, we evaluate the security of several state-of-the-art fraud detection systems by deploying our poisoning attack with different degrees of attacker’s knowledge and attacking strategies. The experimental results show that our attack works even when the attacker has minimal knowledge of the target system. Then, we demonstrate that the proposed countermeasure can mitigate adversarial attacks by reducing the stolen amount of money up to 100%.

Current research in the automotive domain has proven the limitations of the Controller Area Network (CAN) protocol from a security standpoint. Application-layer attacks, which involve the creation of malicious packets, are deemed feasible from remote but can be easily detected by modern Intrusion Detection Systems (IDSs). On the other hand, more recent link-layer attacks are stealthier and possibly more disruptive but require physical access to the bus. In this paper, we present CANflict, a software-only approach that allows reliable manipulation of the CAN bus at the data link layer from an unmodified microcontroller, overcoming the limitations of state-of-the-art works. We demonstrate that it is possible to deploy stealthy CAN link-layer attacks from a remotely compromised ECU, targeting another ECU on the same CAN network. To do this, we exploit the presence of pin conflicts between microcontroller peripherals to craft polyglot frames, which allows an attacker to control the CAN traffic at the bit level and bypass the protocol's rules. We experimentally demonstrate the effectiveness of our approach on high-, mid-, and low-end microcontrollers, and we provide the ground for future research by releasing an extensible tool that can be used to implement our approach on different platforms and to build CAN countermeasures at the data link layer.

Intelligent Transportation Systems (ITSs) technology has advanced during the past years, and it is now used for several applications that require vehicles to exchange real-time data, such as in traffic information management. Traditionally, road traffic information has been collected using on-site sensors. However, crowd-sourcing traffic information from onboard sensors or smartphones has become a viable alternative. State-of-the-art solutions currently follow a centralized model where only the service provider has complete access to the collected traffic data and represent a single point of failure and trust. In this paper, we proposeGOLIATH, a blockchain-based decentralized framework that runs on the In-Vehicle Infotainment (IVI) system to collect real-time information exchanged between the network’s participants. Our approach mitigates the limitations of existing crowd-sourcing centralized solutions by guaranteeing trusted information collection and exchange, fully exploiting the intrinsic distributed nature of vehicles. We demonstrate its feasibility in the context of vehicle positioning and traffic information management. Each vehicle participating in the decentralized network shares its position and neighbors’ ones in the form of a transaction recorded on the ledger, which uses a novel consensus mechanism to validate it. We design the consensus mechanism resilient against a realistic set of adversaries that aim to tamper or disable the communication. We evaluate the proposed framework in a simulated (but realistic) environment, which considers different threats and allows showing its robustness and safety properties.

Automotive security has gained significant traction in the last decade thanks to the development of new connectivity features that have brought the vehicle from an isolated environment to an externally facing domain. Researchers have shown that modern vehicles are vulnerable to multiple types of attacks leveraging remote, direct and indirect physical access, which allow attackers to gain control and affect safety-critical systems. Conversely, Intrusion Detection Systems (IDSs) have been proposed by both industry and academia to identify attacks and anomalous behaviours. In this article, we propose CANnolo, an IDS based on Long Short-Term Memory (LSTM)-autoencoders to identify anomalies in Controller Area Networks (CANs). During a training phase, CANnolo automatically analyzes the CAN streams and builds a model of the legitimate data sequences. Then, it detects anomalies by computing the difference between the reconstructed and the respective real sequences. We experimentally evaluated CANnolo on a set of simulated attacks applied over a real-world dataset. We show that our approach outperforms the state-of-the-art model by improving the detection rate and precision.

Recently, reconstruction-based anomaly detection was proposed as an effective technique to detect attacks in dynamic industrial control networks. Unlike classical network anomaly detectors that observe the network traffic, reconstruction-based detectors operate on the measured sensor data, leveraging physical process models learned a priori.

In the last years, NoSQL databases have grown in popularity due to their easy-to-deploy, reliable, and scalable storage mechanism. While most NoSQL services offer access control mechanisms, their default configurations grant access without any form of authentication, resulting in misconfigurations that may expose data to the Internet, as demonstrated by the recent high-profile data leaks.

The significant growth of banking fraud, fueled by the underground economy of malware, has raised the need for effective detection systems. Therefore, in the last few years, banks have upgraded their security to protect transactions from fraud. State-of-the-art solutions detect fraud as deviations from customers’ spending habits. To the best of our knowledge, almost all existing approaches do not provide an in-depth model’s granularity and security analysis against elusive attacks. In this article, we examine Banksealer, a decision support system for banking fraud analysis that evaluates the influence on detection performance of the granularity at which spending habits are modeled and its security against evasive attacks. First, we compare user-centric modeling, which builds a model for each user, with system-centric modeling, which builds a model for the entire system, from the point of view of detection performance. Then, we assess the robustness of Banksealer against malicious attackers that are aware of the structure of the models in use. To this end, we design and implement a proof-of-concept attack tool that performs mimicry attacks, emulating a sophisticated attacker that cloaks frauds to avoid detection. We experimentally confirm the feasibility of such attacks, their cost, and the effort required by an attacker in order to perform them. In addition, we discuss possible countermeasures. We provide a comprehensive evaluation on a large real-world dataset obtained from one of the largest Italian banks.

Nowadays Information stealers are reaching high levels of sophistication. The number of families and variants observed increased exponentially in the last years. Furthermore, these trojans are sold on underground markets along with automatic frameworks that include web-based administration panels, builders and customization procedures. From a technical point of view such malware is equipped with a functionality, called WebInject, that exploits API hooking techniques to intercept all sensitive data in a browser context and modify web pages on infected hosts. In this paper we propose Prometheus, an automatic system that is able to analyze trojans that base their attack technique on DOM modifications. Prometheus is able to identify the injection operations performed by malware, and generate signatures based on the injection behavior. Furthermore, it is able to extract the WebInject targets by using memory forensic techniques. We evaluated Prometheus against real-world, online websites and a dataset of distinct variants of financial trojans. In our experiments we show that our approach correctly recognizes known variants of WebInject-based malware and successfully extracts the WebInject targets.