Search

Smart security devices, such as smart locks, smart cameras, and smart intruder alarms are increasingly popular with users due to the enhanced convenience and new features that they offer. A significant part of this convenience is provided by the device’s companion smartphone app. Information on whether secure and ethical development practices have been used in the creation of these applications is unavailable to the end user. As this work shows, this means that users are impacted both by potential third-party attackers that aim to compromise their device, and more subtle threats introduced by developers, who may track their use of their devices and illegally collect data that violate users’ privacy. Our results suggest that users of every application tested are susceptible to at least one potential commonly found vulnerability regardless of whether their device is offered by a known brand name or a lesser-known manufacturer. We present an overview of the most common vulnerabilities found in the scanned code and discuss the shortcomings of state-of-the-art automated scanners when looking at less structured programming languages such as C and C++. Finally, we also discuss potential methods for mitigation, and provide recommendations for developers to follow with respect to secure coding practices.

Software-Defined Networking (SDN), which is used in Industrial Internet of Things, uses a controller as its “network brain” located at the control plane. This uniquely distinguishes it from the traditional networking paradigms because it provides a global view of the entire network. In SDN, the controller can become a single point of failure, which may cause the whole network service to be compromised. Also, data packet transmission between controllers and switches could be impaired by natural disasters, causing hardware malfunctioning or Distributed Denial of Service (DDoS) attacks. Thus, SDN controllers are vulnerable to both hardware and software failures. To overcome this single point of failure in SDN, this paper proposes an attack-aware logical link assignment (AALLA) mathematical model with the ultimate aim of restoring the SDN network by using logical link assignment from switches to the cluster (backup) controllers. We formulate the AALLA model in integer linear programming (ILP), which restores the disrupted SDN network availability by assigning the logical links to the cluster (backup) controllers. More precisely, given a set of switches that are managed by the controller(s), this model simultaneously determines the optimal cost for controllers, links, and switches.

As the internet continues to be populated with new devices and emerging technologies, the attack surface grows exponentially. Technology is shifting towards a profit-driven Internet of Things market where security is an afterthought. Traditional defending approaches are no longer sufficient to detect both known and unknown attacks to high accuracy. Machine learning intrusion detection systems have proven their success in identifying unknown attacks with high precision. Nevertheless, machine learning models are also vulnerable to attacks. Adversarial examples can be used to evaluate the robustness of a designed model before it is deployed. Further, using adversarial examples is critical to creating a robust model designed for an adversarial environment. Our work evaluates both traditional machine learning and deep learning models’ robustness using the Bot-IoT dataset. Our methodology included two main approaches. First, label poisoning, used to cause incorrect classification by the model. Second, the fast gradient sign method, used to evade detection measures. The experiments demonstrated that an attacker could manipulate or circumvent detection with significant probability.

The proliferation of social media platforms changed the way people interact online. However, engagement with social media comes with a price, the users’ privacy. Breaches of users’ privacy, such as the Cambridge Analytica scandal, can reveal how the users’ data can be weaponized in political campaigns, which many times trigger hate speech and anti-immigration views. Hate speech detection is a challenging task due to the different sources of hate that can have an impact on the language used, as well as the lack of relevant annotated data. To tackle this, we collected and manually annotated an immigration-related dataset of publicly available Tweets in UK, US, and Canadian English. In an empirical study, we explored anti-immigration speech detection utilizing various language features (word n-grams, character n-grams) and measured their impact on a number of trained classifiers. Our work demonstrates that using word n-grams results in higher precision, recall, and f-score as compared to character n-grams. Finally, we discuss the implications of these results for future work on hate-speech detection and social media data analysis in general.

The distributed nature of mobile ad hoc networks (MANETs) presents security challenges and vulnerabilities which sometimes lead to several forms of attacks. To improve the security in MANETs, reputation and trust management systems (RTMS) have been developed to mitigate some attacks and threats arising from abnormal behaviours of nodes in networks. Generally, most reputation and trust systems in MANETs focus mainly on penalising uncooperative network nodes. It is a known fact that nodes in MANETs have limited energy resources and as such, the continuous collaboration of cooperative nodes will lead to energy exhaustion. This paper develops and evaluates a robust Dirichlet reputation and trust management system which measures and models the reputation and trust of nodes in the network, and it incorporates candour into the mode of operations of the RTMS without undermining network security. The proposed RTMS employs Dirichlet probability distribution in modelling the individual reputation of nodes and the trust of each node is computed based on the node’s actual network performance and the accuracy of the second-hand reputations it gives about other nodes. The paper also presents a novel candour two-dimensional trustworthiness evaluation technique that categorises the behaviours of nodes based on their evaluated total reputation and trust values. The evaluation and analyses of some of the simulated behaviours of nodes in the deployed MANETs show that the candour two-dimensional trustworthiness evaluation technique is an effective technique that encourages and caters to nodes that continuously contribute to the network despite the reduction in their energy levels.

As the usage of the Web increases, so do the threats an everyday user faces. One of the most pervasive threats a Web user faces is tracking, which enables an entity to gain unauthorized access to the user's personal data. Through the years, many client storage technologies, such as cookies, have been used for this purpose and have been extensively studied in the literature. The focus of this paper is on three newer client storage mechanisms, namely, Web Storage, Web SQL Database, and Indexed Database API. Initially, a large-scale analysis of their usage on the Web is conducted to appraise their usage in the wild. Then, this paper examines the extent that they are used for tracking purposes. The results suggest that Web Storage is the most used among the three technologies. More importantly, to the best of our knowledge, this paper is the first to suggest Web tracking as the main use case of these technologies. Motivated by these results, this paper examines whether popular desktop and mobile browsers protect their users from tracking mechanisms that use Web Storage, Web SQL Database, and Indexed Database. Our results uncover many cases where the relevant security controls are ineffective, thus making it virtually impossible for certain users to avoid tracking.

Nowadays, the majority of everyday computing devices, irrespective of their size and operating system, allow access to information and online services through web browsers. However, the pervasiveness of web browsing in our daily life does not come without security risks. This widespread practice of web browsing in combination with web users’ low situational awareness against cyber attacks, exposes them to a variety of threats, such as phishing, malware and profiling. Phishing attacks can compromise a target, individual or enterprise, through social interaction alone. Moreover, in the current threat landscape phishing attacks typically serve as an attack vector or initial step in a more complex campaign. To make matters worse, past work has demonstrated the inability of denylists, which are the default phishing countermeasure, to protect users from the dynamic nature of phishing URLs. In this context, our work uses supervised machine learning to block phishing attacks, based on a novel combination of features that are extracted solely from the URL. We evaluate our performance over time with a dataset which consists of active phishing attacks and compare it with Google Safe Browsing (GSB), i.e., the default security control in most popular web browsers. We find that our work outperforms GSB in all of our experiments, as well as performs well even against phishing URLs which are active one year after our model’s training.

In 2019, the majority of companies used at least one cloud computing service and it is expected that by the end of 2021, cloud data centres will process 94% of workloads. The financial and operational advantages of moving IT infrastructure to specialised cloud providers are clearly compelling. However, with such volumes of private and personal data being stored in cloud computing infrastructures, security concerns have risen. Motivated to monitor and analyze adversarial activities, we deploy multiple honeypots on the popular cloud providers, namely Amazon Web Services (AWS), Google Cloud Platform (GCP) and Microsoft Azure, and operate them in multiple regions. Logs were collected over a period of three weeks in May 2020 and then comparatively analysed, evaluated and visualised. Our work revealed heterogeneous attackers’ activity on each cloud provider, both when one considers the volume and origin of attacks, as well as the targeted services and vulnerabilities. Our results highlight the attempt of threat actors to abuse popular services, which were widely used during the COVID-19 pandemic for remote working, such as remote desktop sharing. Furthermore, the attacks seem to exit not only from countries that are commonly found to be the source of attacks, such as China, Russia and the United States, but also from uncommon ones such as Vietnam, India and Venezuela. Our results provide insights on the adversarial activity during our experiments, which can be used to inform the Situational Awareness operations of an organisation.

Social media has become very popular and important in people’s lives, as personal ideas, beliefs and opinions are expressed and shared through them. Unfortunately, social networks, and specifically Twitter, suffer from massive existence and perpetual creation of fake users. Their goal is to deceive other users employing various methods, or even create a stream of fake news and opinions in order to influence an idea upon a specific subject, thus impairing the platform’s integrity. As such, machine learning techniques have been widely used in social networks to address this type of threat by automatically identifying fake accounts. Nonetheless, threat actors update their arsenal and launch a range of sophisticated attacks to undermine this detection procedure, either during the training or test phase, rendering machine learning algorithms vulnerable to adversarial attacks. Our work examines the propagation of adversarial attacks in machine learning based detection for fake Twitter accounts, which is based on AdaBoost. Moreover, we propose and evaluate the use of k-NN as a countermeasure to remedy the effects of the adversarial attacks that we have implemented.

Phishing is one of the most common threats that users face while browsing the web. In the current threat landscape, a targeted phishing attack (i.e., spear phishing) often constitutes the first action of a threat actor during an intrusion campaign. To tackle this threat, many data-driven approaches have been proposed, which mostly rely on the use of supervised machine learning under a single-layer approach. However, such approaches are resource-demanding and, thus, their deployment in production environments is infeasible. Moreover, most previous works utilise a feature set that can be easily tampered with by adversaries. In this paper, we investigate the use of a multi-layered detection framework in which a potential phishing domain is classified multiple times by models using different feature sets. In our work, an additional classification takes place only when the initial one scores below a predefined confidence level, which is set by the system owner. We demonstrate our approach by implementing a two-layered detection system, which uses supervised machine learning to identify phishing attacks. We evaluate our system with a dataset consisting of active phishing attacks and find that its performance is comparable to the state of the art.