Anomaly-based ML-NIDS (A-NIDS) model normal network behavior from benign data and classify deviations from this baseline as anomalies, theoretically enabling the detection of evolving attack variants without labeled attack data. The ability of A-NIDS to generalize critically depends on the quality of the feature space representing network behavior. However, the requirement for feature spaces that encode attack-relevant semantics has received little attention and remains poorly understood. As a consequence, these systems still struggle to meet practical operational constraints (low false positive rates without compromising detection performance and generalization to attack variants). We identify two limitations in the current feature spaces. First, Out-of-Dimension Blindness, where features do not capture essential attack mechanism properties. Second, Attack Strategy Aggregation Failure, where features cannot encode composite attack behaviors. Moreover, we demonstrate that two SotA data-driven generalization frameworks (based on incremental and contrastive learning) cannot compensate for these feature-level shortcomings. To bridge this gap, we present KnowML, a framework that encodes attack domain knowledge directly into the feature space. For each attack family, our method employs LLMs to construct a corresponding Knowledge Graph (KG) from attack implementations. Symbolic reasoning is then applied over the KG to enumerate potential attack strategies and their compositions. The resulting Knowledge-Augmented Feature Space enables effective generalization even when trained exclusively on benign traffic, a capability beyond current approaches. Systematic empirical evaluations show that KnowML achieves up to 99% detection rates while maintaining false positive rates at or below 0.0137%, substantially outperforming contemporary feature-based baselines across diverse attack variants.

The scarcity of data and the high complexity of Advanced Persistent Threats (APTs) attacks have created challenges in comprehending their behavior and hindered the exploration of effective detection techniques. To create an effective APT detection strategy, it is important to examine the Tactics, Techniques, and Procedures (TTPs) that have been reported by the industry. These TTPs can be difficult to classify as either malicious or legitimate. When developing an approach for the next generation of network intrusion detection systems (NIDS), it is necessary to take into account the specific context of the attack explained in this paper. In this study, we select 33 APT campaigns based on the fair distribution over the past 22 years to observe the evolution of APTs over time. We focus on their evasion techniques and how they stay undetected for months or years. We found that APTs cannot continue their operations without C& C servers, which are mostly addressed by Domain Name System (DNS). We identify several TTPs used for DNS, such as Dynamic DNS, typosquatting, and TLD squatting. The next step for APT operators is to start communicating with a victim. We found that the most popular protocol to deploy evasion techniques is using HTTP(S) with 81% of APT campaigns. HTTP(S) can evade firewall filtering and pose as legitimate web-based traffic. DNS protocol is also widely used by 45% of APTs for DNS resolution and tunneling. We also investigated the excessive use of fallback channels to evade NIDS that rely on volume-based features. We identify that 60.6% of APT campaigns split their traffic over multiple IP addresses. We also highlight other TTPs that can also be equipped during a campaign, such as non-application protocols or data obfuscation, which are frequently found in our analysis. In conclusion, we outline a roadmap for the future development of next-generation APT detection systems. Our study highlights the necessity of innovative solutions to counter evolving attack patterns, improve threat visibility, and reinforce proactive defense strategies.

REST APIs have become key components of web services. However, they often contain logic flaws resulting in server side errors or security vulnerabilities. HTTP requests are used as test cases to find and mitigate such issues. Existing methods to modify requests, including those using deep learning, suffer from limited performance and precision, relying on undirected search or making limited usage of the contextual information. In this paper we propose APIRL, a fully automated deep reinforcement learning tool for testing REST APIs. A key novelty of our approach is the use of feedback from a transformer module pre-trained on JSON-structured data, akin to that used in API responses. This allows APIRL to learn the subtleties relating to test outcomes, and generalise to unseen API endpoints. We show APIRL can find significantly more bugs than the state-of-the-art in real world REST APIs while minimising the number of required test cases. We also study how reward functions, and other key design choices, affect learnt policies with a thorough ablation study.

Cyberattack investigations are crucial for understanding the Tactics, Techniques, and Procedures of adversaries, but they face increasing challenges due to the complexity, scale, and volume of modern cyber incidents. Current approaches, such as heuristic-based and learning-based methods, struggle with scalability and reliance on labeled data, leading to difficulties in adapting to new threats. In this paper, we present Clouseau,a hierarchical multi-agent approach that leverages the reasoning capabilities of Large Language Models (LLMs) to autonomously investigate cyberattacks from a single Point-Of-Interest while requiring neither prior training nor predefined heuristics. We evaluated Clouseauon 21 diverse attack scenarios, including complex cases from DARPA's OpTC engagements and Advanced Persistent Threat (APT) attack scenarios from the ATLAS dataset. In single-host settings, Clouseauachieves an average F1 score of 99.78%, surpassing strong baselines by more than 33%. To demonstrate its broad applicability, we tested Clouseau with both proprietary and open-weight LLMs, achieving strong performance in both cases, thereby enabling deployment in private environments where access to proprietary models is restricted.

Deep learning approaches have achieved remarkable performance in malware classification and detection. However, their success relies on the availability of large, accurately labeled datasets: a critical yet challenging requirement in the malware domain. In practice, most malware datasets are automatically labeled using outputs from antivirus engines, a process that often introduces significant label noise. Such imperfections can severely degrade the performance and generalizability of deep learning models.

Deep learning techniques are increasingly being incorporated into NIDS. However, the evaluation of such deep learning models often assumes static data distributions and overlooks the effects of randomness and environmental variation. As a result, the reported performance may not reflect the NIDS behaviour during real-world deployment. This paper investigates the impact of stochastic and environmental factors on the evaluation of deep learning models for NIDS, with a focus on shift-aware models that detect and adapt to data shift, representing state-of-the-art systems for long-term deployment. We examine two baselines under controlled variations to analyse the impact of each factor on the reproducibility and fairness of the results, revealing that the F1 score can vary largely due to these, even minor, variations. All of the explored factors affect the reproducibility of the results, and some can significantly skew performance. Based on our findings, we provide practical recommendations to support reproducible and fair evaluations of deep learning-based NIDS systems.

Anomaly-based intrusion detection systems are tasked with identifying deviations from established benign network behaviors, assuming such deviations to be indicators of malicious intent. Deep AutoEncoders (DAEs) have become increasingly popular in these systems due to their exceptional ability to model benign behavior with high accuracy, particularly in static, offline settings where the network’s benign activity pattern is presumed to remain constant. However, this static approach becomes less effective as network behavior naturally evolves, leading to challenges in distinguishing new, benign activities from genuine threats. This evolution raises a critical question: How can we enhance offline DAEs to accurately identify threats while avoiding false alarms caused by benign behavior changes?

This paper presents VulBERTa, a deep learning approach to detect security vulnerabilities in source code. Our approach pre-trains a RoBERTa model with a custom tokenisation pipeline on real-world code from open-source C/C++ projects. The model learns a deep knowledge representation of the code syntax and semantics, which we leverage to train vulnerability detection classifiers. We evaluate our approach on binary and multi-class vulnerability detection tasks across several datasets (Vuldeepecker, Draper, REVEAL and muVuldeepecker) and benchmarks (CodeXGLUE and D2A). The evaluation results show that VulBERTa achieves state-of-the-art performance and outperforms existing approaches across different datasets, despite its conceptual simplicity, and limited cost in terms of size of training data and number of model parameters.

Web application vulnerabilities are an ongoing problem that current black-box techniques and scanners do not entirely solve, suffering in particular from a lack of payload diversity that prevents them from capturing the long tail of vulnerabilities caused by uncommon sanitisation mistakes.In order to increase the diversity of payloads that can be automatically generated in a black-box fashion, we develop a hierarchical reinforcement learning approach where agents focus separately on the tasks of escaping the current context, and evading sanitisation. We implement this in an end-to-end prototype we call HAXSS.We compare our approach against a number of state-of-the-art black-box scanners on a new micro-benchmark for XSS payload generation, and on a macro-benchmark of established vulnerable web applications. HAXSS outperforms the other scanners on both benchmarks, identifying 131 vulnerabilities (a 20% improvement over the closest scanner), reporting 0 false positives. Finally, we demonstrate that our approach is practically useful, as HAXSS re-discovers 4 existing CVEs and discovers 5 new CVEs in 3 production-grade web applications.

Neural networks are increasingly used for intrusion detection on industrial control systems (ICS). With neural networks being vulnerable to adversarial examples, attackers who wish to cause damage to an ICS can attempt to hide their attacks from detection by using adversarial example techniques. In this work we address the domain specific challenges of constructing such attacks against autoregressive based intrusion detection systems (IDS) in an ICS setting. We model an attacker that can compromise a subset of sensors in a ICS which has a LSTM based IDS. The attacker manipulates the data sent to the IDS, and seeks to hide the presence of real cyber-physical attacks occurring in the ICS. We evaluate our adversarial attack methodology on the Secure Water Treatment system when examining solely continuous data, and on data containing a mixture of discrete and continuous variables. In the continuous data domain our attack successfully hides the cyber-physical attacks requiring 2.87 out of 12 monitored sensors to be compromised on average. With both discrete and continuous data our attack required, on average, 3.74 out of 26 monitored sensors to be compromised.

Machine learning systems have had enormous success in a wide range of fields from computer vision, natural language processing, and anomaly detection. However, such systems are vulnerable to attackers who can cause deliberate misclassification by introducing small perturbations. With machine learning systems being proposed for cyber attack detection such attackers are cause for serious concern. Despite this the vast majority of adversarial machine learning security research is focused on the image domain. This work gives a brief overview of adversarial machine learning and machine learning used in cyber attack detection and suggests key differences between the traditional image domain of adversarial machine learning and the cyber domain. Finally we show an adversarial machine learning attack on an industrial control system.

Monitoring systems are essential to understand and control the behaviour of systems and networks. Cyber-physical systems (CPS) are particularly delicate under that perspective since they involve real-time constraints and physical phenomena that are not usually considered in common IT solutions. Therefore, there is a need for publicly available monitoring tools able to contemplate these aspects. In this poster/demo, we present our initiative, called CPS-MT, towards a versatile, real-time CPS monitoring tool, with a particular focus on security research. We first present its architecture and main components, followed by a MiniCPS-based case study. We also describe a performance analysis and preliminary results. During the demo, we will discuss CPS-MT's capabilities and limitations for security applications.

The security of the client side of a web application relies on browser features such as cookies, the same-origin policy and HTTPS. As the client side grows increasingly powerful and sophisticated, browser vendors have stepped up their offering of security mechanisms which can be leveraged to protect it. These are often introduced experimentally and informally and, as adoption increases, gradually become standardised (e.g., CSP, CORS and HSTS). Considering the diverse landscape of browser vendors, releases, and customised versions for mobile and embedded devices, there is a compelling need for a systematic assessment of browser security. We present BrowserAudit, a tool for testing that a deployed browser enforces the guarantees implied by the main standardised and experimental security mechanisms. It includes more than 400 fully-automated tests that exercise a broad range of security features, helping web users, application developers and security researchers to make an informed security assessment of a deployed browser. We validate BrowserAudit by discovering both fresh and known security-related bugs in major browsers.

JavaScript is the most widely used web language for client-side applications. Whilst the development of JavaScript was initially just led by implementation, there is now increasing momentum behind the ECMA standardisation process. The time is ripe for a formal, mechanised specification of JavaScript, to clarify ambiguities in the ECMA standards, to serve as a trusted reference for high-level language compilation and JavaScript implementations, and to provide a platform for high-assurance proofs of language properties.

Social sign-on and social sharing are becoming an ever more popular feature of web applications. This success is largely due to the APIs and support offered by prominent social networks, such as Facebook, Twitter and Google, on the basis of new open standards such as the OAuth 2.0 authorization pro tocol. A formal analysis of these protocols must account for malicious websites and common web application vulnerabilities, such as cross-site request forgery and open redirectors. We model several configurations of the OAuth 2.0 protocol in the applied pi-calculus and verify them using ProVerif. Our models rely on WebSpi, a new library for modeling web applications and web-based attackers that is designed to help discover concrete attacks on websites. To ease the task of writing formal models in our framework, we present a model extraction tool that automatically translates programs written in subsets of PHP and JavaScript to the applied pi-calculus. Our approach is validated by finding dozens of previously unknown vulnerabilities in popular websites such as Yahoo and WordPress, when they connect to social networks such as Twitter and Facebook.

Social sign-on and social sharing are becoming an ever more popular feature of web applications. This success is largely due to the APIs and support offered by prominent social networks, such as Facebook, Twitter, and Google, on the basis of new open standards such as the OAuth 2.0 authorization protocol. A formal analysis of these protocols must account for malicious websites and common web application vulnerabilities, such as cross-site request forgery and open redirectors. We model several configurations of the OAuth 2.0 protocol in the applied pi-calculus and verify them using ProVerif. Our models rely on WebSpi, a new library for modeling web applications and web-based attackers that is designed to help discover concrete website attacks. Our approach is validated by finding dozens of previously unknown vulnerabilities in popular websites such as Yahoo and Word Press, when they connect to social networks such as Twitter and Facebook.

JavaScript has become the most widely used language for client-side web programming. The dynamic nature of JavaScript makes understanding its code notoriously difficult, leading to buggy programs and a lack of adequate static-analysis tools. We believe that logical reasoning has much to offer JavaScript: a simple description of program behaviour, a clear understanding of module boundaries, and the ability to verify security contracts. We introduce a program logic for reasoning about a broad subset of JavaScript, including challenging features such as prototype inheritance and "with". We adapt ideas from separation logic to provide tractable reasoning about JavaScript code: reasoning about easy programs is easy; reasoning about hard programs is possible. We prove a strong soundness result. All libraries written in our subset and proved correct with respect to their specifications will be well-behaved, even when called by arbitrary JavaScript code.

We present the design and implementation of a typechecker for verifying security properties of the source code of cryptographic protocols and access control mechanisms. The underlying type theory is a λ-calculus equipped with refinement types for expressing pre- and post-conditions within first-order logic. We derive formal cryptographic primitives and represent active adversaries within the type theory. Well-typed programs enjoy assertion-based security properties, with respect to a realistic threat model including key compromise. The implementation amounts to an enhanced typechecker for the general-purpose functional language F # ; typechecking generates verification conditions that are passed to an SMT solver. We describe a series of checked examples. This is the first tool to verify authentication properties of cryptographic protocols by typechecking their source code.

A growing number of current web sites combine active content (applications) from untrusted sources, as in so-called mashups. The object-capability model provides an appealing approach for isolating untrusted content: if separate applications are provided disjoint capabilities, a sound object capability framework should prevent untrusted applications from interfering with each other, without preventing interaction with the user or the hosting page. In developing language-based foundations for isolation proofs based on object-capability concepts, we identify a more general notion of authority safety that also implies resource isolation. After proving that capability safety implies authority safety, we show the applicability of our framework for a specific class of mashups. In addition to proving that a JavaScript subset based on Google Caja is capability safe, we prove that a more expressive subset of JavaScript is authority safe, even though it is not based on the object-capability model.

Web sites that incorporate untrusted content may use browser- or language-based methods to keep such content from maliciously altering pages, stealing sensitive information, or causing other harm. We study language-based methods for filtering and rewriting JavaScript code, using Yahoo! ADSafe and Facebook FBJS as motivating examples. We explain the core problems by describing previously unknown vulnerabilities and subtleties, and develop a foundation for improved solutions based on an operational semantics of the full ECMA-262 language. We also discuss how to apply our analysis to address the JavaScript isolation problems we discovered.

We present the design and implementation of a typechecker for verifying security properties of the source code of cryptographic protocols and access control mechanisms. The underlying type theory is a λ-calculus equipped with refinement types for expressing pre- and post-conditions within first-order logic. We derive formal cryptographic primitives and represent active adversaries within the type theory. Well-typed programs enjoy assertion-based security properties, with respect to a realistic threat model including key compromise. The implementation amounts to an enhanced typechecker for the general purpose functional language F#; typechecking generates verification conditions that are passed to an SMT solver. We describe a series of checked examples. This is the first tool to verify authentication properties of cryptographic protocols by typechecking their source code.

We consider the problem of statically verifying the conformance of the code of a system to an explicit authorization policy. In a distributed setting, some part of the system may be compromised, that is, some nodes of the system and their security credentials may be under the control of an attacker. To help predict and bound the impact of such partial compromise, we advocate logic-based policies that explicitly record dependencies between principals. We propose a conformance criterion, safety despite compromised principals, such that an invalid authorization decision at an uncompromised node can arise only when nodes on which the decision logically depends are compromised. We formalize this criterion in the setting of a process calculus, and present a verification technique based on a type system. Hence, we can verify policy conformance of code that uses a wide range of the security mechanisms found in distributed systems, ranging from secure channels down to cryptographic primitives, including encryption and public-key signatures.

Distributed systems and applications are often expected to enforce high-level authorization policies. To this end, the code for these systems relies on lower-level security mechanisms such as digital signatures, local ACLs, and encrypted communications. In principle, authorization specifications can be separated from code and carefully audited. Logic programs in particular can express policies in a simple, abstract manner. We consider the problem of checking whether a distributed implementation based on communication channels and cryptography complies with a logical authorization policy. We formalize authorization policies and their connection to code by embedding logical predicates and claims within a process calculus. We formulate policy compliance operationally by composing a process model of the distributed system with an arbitrary opponent process. Moreover, we propose a dependent type system for verifying policy compliance of implementation code. Using Datalog as an authorization logic, we show how to type several examples using policies and present a general schema for compiling policies.