Learning from cyber incidents is crucial for organisations to enhance their cyber resilience and effectively respond to evolving threats. This study employs neo-institutional and organisational learning theories to examine how organisations learn from incidents and gain insights into the challenges they face. Drawing on qualitative research methods, interviews were conducted with 34 security practitioners from organisations operating in the UK spanning a range of industries. The findings highlight the importance of consciously evaluating learning practices and creating a culture of openness to hear about incidents from employees, customers and suppliers. Deciding which incidents to learn from, as well as who should participate in the learning process, emerged as critical considerations. Overcoming defensiveness and addressing systemic causes were recognised as barriers to effective learning. The study emphasises the need to assess the value and impact of identified lessons and to avoid superficial reviews that treat symptoms rather than underlying causes to improve resilience. While progress has been made in learning from incidents, further enhancements are needed. Practical recommendations have been proposed to suggest how organisations may gain valuable insights for maximising the benefits derived from incident learning. This research contributes to the existing knowledge on organisational learning and informs future studies exploring the social and political influences on the learning process. By considering the suggested recommendations, organisations may strengthen their cyber security, foster a culture of continuous improvement, and respond effectively to the dynamic cyber security landscape.

Abstract Manually labelling datasets for training violence detection systems is time-consuming, expensive, and labor-intensive. Mind wandering, boredom, and short attention span can also cause labelling errors. Moreover, collecting and distributing sensitive images containing violence has ethical implications. Automation is the future for labelling sensitive image datasets. Deep labeller is a two-stage Deep Learning (DL) method that uses pre-trained DL object detection methods on MS-COCO for automatic labelling. The Deep Labeller method labels violent and nonviolent images in WVD and USI. In stage 1, WVD generates weak labels using synthetic images. In stage 2, the Deep labeller method is retrained on weak labels. USI dataset is used to test our method on real-world violence. Deep labeller generated weak and strong labels with an IoU of 0.80036 in stage 1 and 0.95 in stage 2 on the WVD. Automatically generated labels. To test our method’s generalisation power, violent and nonviolent image labels on USI dataset had a mean IoU of 0.7450.

Cyber security incidents are now prevalent in many organisations. Arguably, those who can learn from security incidents and address the underlying causes will reduce the prevalence of similar ones in the future. This research provides a new examination of how organisations learn from incidents by systematically reviewing academic research on organisational learning from cyber security incidents and identifying further research needed in this area. To do this, it considers three research questions: what research has been conducted on learning from cyber security incidents, what learning practices in organisations have been found by research and what improvements have been recommended, and what further research is needed as organisations learn from such incidents. Using the PRISMA method, a total of 3,986 articles were extracted and, from these, a relevant set of 30 were selected for analysis to map the body of research, and to identify future research avenues. Despite learning lessons being recommended by both researchers and industry standards, our findings suggest that this advice is not being fully adopted by organisations. Importantly, these studies have found inadequate participation in learning activities, with superficial causal investigations, scarce effort on ensuring lessons are implemented and no evaluation of whether the actions taken actually reduce future security incidents. More research is needed to understand the right level and which learning practices to invest in for the greatest impact. For practitioners, this review discusses the essential elements of an effective process to learn from incidents. This review provides academics with a novel synthesis of the research undertaken on this topic, enabling them to incorporate the significant findings into their work and potentially explore the research agenda suggested.

Cyber security is a topic of increasing relevance in relation to industrial networks. The higher intensity and intelligent use of data pushed by smart technology (Industry 4.0) together with an augmented integration between the operational technology (production) and the information technology (business) parts of the network have considerably raised the level of vulnerabilities. On the other hand, many industrial facilities still use serial networks as underlying communication system, and they are notoriously limited from a cyber security perspective since protection mechanisms available for ТСР/IР communication do not apply. Therefore, an attacker gaining access to a serial network can easily control the industrial components, potentially causing catastrophic incidents, jeopardizing assets and human lives. This study proposes a framework to act as an anomaly detection system (ADS) for industrial serial networks. It has three ingredients: an unsupervised К-means component to analyse message content, a knowledge-based Expert System component to analyse message metadata, and a voting process to generate alerts for security incidents, anomalous states, and faults. The framework was evaluated using the Proflbus-DP, a network simulator which implements a serial bus system. Results for the simulated traffic were promising: 99.90% for accuracy, 99,64% for precision, and 99.28% for F1-Score. They indicate feasibility of the framework applied to serial-based industrial networks.

Cloud computing has gained popularity due to its efficiency, robustness and cost effectiveness. Carrying out digital forensic investigations in the cloud is currently a relevant and open issue. The root of this issue is the fact that servers cannot be physically accessed, coupled with the dynamic and distributed nature of cloud computing with regards to data processing and storage. This renders traditional methods of evidence collection impractical. The use of provenance data in cloud forensics is critical as it provides forensic investigators with data history in terms of people, entities and activities involved in producing related data objects. Therefore, cloud forensics requires effective provenance collection mechanisms. This paper provides an overview of current provenance challenges in cloud computing and identifies limitations of current provenance collection mechanisms. Recommendations for additional research in digital provenance for cloud forensics are also presented.

The utility of Behavioural Evidence Analysis (BEA) has gained attention in the field of Digital Forensics in recent years. It has been recognized that, along with technical examination of digital evidence, it is important to learn as much as possible about the individuals behind an offence, the victim (s) and the dynamics of a crime. This can assist the investigator in producing a more accurate and complete reconstruction of the crime, in interpreting associated digital evidence, and with the description of investigative findings. Despite these potential benefits, the literature shows limited use of BEA for the investigation of cases of the possession and dissemination of Sexually Exploitative Imagery of Children (SEIC). This paper represents a step towards filling this gap. It reports on the forensic analysis of 15 SEIC cases involving P2P file sharing networks, obtained from the Dubai Police. Results confirmed the predicted benefits and indicate that BEA can assist digital forensic practitioners and prosecutors.

No abstract available.

When showing that a software system meets certain security requirements, it is often necessary to work with formal and informal descriptions of the system behavior, vulnerabilities, and threats from potential attackers. In earlier work, Haley et al. [1] showed structured argumentation could deal with such mixed descriptions. However, incomplete and uncertain information, and limited resources force practitioners to settle for good-enough security. To deal with these conditions of practice, we extend the method of Haley et al. with risk assessment. The proposed method, RISA (RIsk assessment in Security Argumentation), uses public catalogs of security expertise to support the risk assessment, and to guide the security argumentation in identifying rebuttals and mitigations for security requirements satisfaction. We illustrate RISA with a realistic example of PIN Entry Device.

When software systems are verified against security requirements, formal and informal arguments provide a structure for organizing the software artifacts. Our recent work on the evolution of security-critical software systems demonstrates that our argumentation technique is useful in limiting the scope of change and in identifying changes to security properties. In support of this work, we have developed OpenArgue, a tool for syntax checking, visualizing, formalizing, and reasoning about incremental arguments. OpenArgue has been integrated with requirements engineering tools for Problem Frames and i∗, and applied to an Air Traffic Management (ATM) case study.

Increasingly, organizations collaborate with other organizations in value webs with various arrangements, such as outsourcing, partnering, joint ventures, or subcontracting. As the Jericho Forum (an industry consortium of the Open Group) observed, in all these forms of collaboration, the boundaries between organizations become permeable and, as a consequence, insiders and outsiders can no longer be neatly separated using the notion of a perimeter. Such organizational arrangements have security implications because individuals from the value web are neither outsiders nor completely insiders. To address this phenomenon this paper proposes a third set of individuals, called External Insiders. External insiders add challenges to the already known insider threat problem because, unlike outsiders, external insiders have granted access and are trusted; and, unlike traditional insiders, external insiders are not subjected to as many internal controls enforced by the organization for which they are external insiders. In fact, external insiders are part of two or more organizational control structures, and business-to-business contracts are often insufficiently detailed to establish security requirements at the level of granularity needed to counter the threat they pose.

Attack graphs are an important support for assessment and subsequent improvement of network security. They reveal possible paths an attacker can take to break through security perimeters and traverse a network to reach valuable assets deep inside the network. Although scalability is no longer the main issue, Attack Graphs still have some problems that make them less useful in practice. First, Attack Graphs remain difficult to relate to the network topology. Second, Attack Graphs traditionally only consider the exploitation of vulnerable hosts. Third, Attack Graphs do not rely on automatic identification of potential attack targets. We address these gaps in our MsAMS (Multi-step Attack Modelling and Simulation) tool, based on Mobile Ambients. The tool not only allows the modelling of more static aspects of the network, such as the network topology, but also the dynamics of network attacks. In addition to Mobile Ambients, we use the pagerank algorithm to determine targets and hub scores produced by the HITS (Hypertext Induced Topic Search) algorithm to guide the simulation of an attacker searching for targets.

Security management is about calculated risk and requires continuous evaluation to ensure cost, time and resource effectiveness. Parts of which is to make future-oriented, cost-benefit investments in security. Security investments must adhere to healthy business principles where both security and financial aspects play an important role. Information on the current and potential risk level is essential to successfully trade-off security and financial aspects. Risk level is the combination of the frequency and impact of a potential unwanted event, often referred to as a security threat or misuse. The paper presents a risk level estimation model that derives risk level as a conditional probability over frequency and impact estimates. The frequency and impact estimates are derived from a set of attributes specified in the Common Vulnerability Scoring System (CVSS). The model works on the level of vulnerabilities (just as the CVSS) and is able to compose vulnerabilities into service levels. The service levels define the potential risk levels and are modelled as a Markov process, which are then used to predict the risk level at a particular time.

This paper presents an evolutionary approach for learning attack specifications that describe attack scenarios. The objective is to find vulnerabilities in computer networks which minimise the cost of an attack with maximum impact. Although we focus on insider threat, the proposed approach applies to networks in general, including social networks and computer grid.