Vasilios Mavroudis

This position article maps large language model (LLM) capabilities onto the MITRE ATT&CK and NIST Cybersecurity Frameworks. While LLMs exhibit fluency and versatility across cybersecurity tasks, they remain fragile in complex, high-stakes settings due to limited context retention, hallucinations, and reasoning gaps.

In the ongoing network cybersecurity arms race, the defenders face a significant disadvantage as they must detect and counteract every attack. Conversely, the attacker only needs to succeed once to achieve their goal. To balance the odds, Autonomous Cyber Network Defence (ACND) employs autonomous agents for proactive and intelligent cyber-attack response. This article surveys the state-of-the-art of Autonomous Blue and Red Teaming agents, as well as cyber operations environments. We begin by presenting a detailed set of criteria for ACND algorithms and systems that evaluate the preparedness of integrating autonomous agents into real-world networked environments. Our analysis identifies critical research gaps and challenges within the ACND landscape, including issues of autonomous agent explainability, continuous learning capability under evolving threats, and the development of realistic agent training environments. Based on these insights, we discuss promising research directions and open challenges that need to be addressed for the deployment of ACND agents in real-world networks.

LangChain is a rapidly emerging framework that offers a versatile and modular approach to developing applications powered by large language models (LLMs). By leveraging LangChain, developers can simplify complex stages of the application lifecycle—such as development, productionization, and deployment—making it easier to build scalable, stateful, and contextually aware applications. It provides tools for handling chat models, integrating retrieval-augmented generation (RAG), and offering secure API interactions. With LangChain, rapid deployment of sophisticated LLM solutions across diverse domains becomes feasible. However, despite its strengths, LangChain's emphasis on modularity and integration introduces complexities and potential security concerns that warrant critical examination. This paper provides an in-depth analysis of LangChain's architecture and core components, including LangGraph, LangServe, and LangSmith. We explore how the framework facilitates the development of LLM applications, discuss its applications across multiple domains, and critically evaluate its limitations in terms of usability, security, and scalability. By offering valuable insights into both the capabilities and challenges of LangChain, this paper serves as a key resource for developers and researchers interested in leveraging LangChain for innovative and secure LLM-powered applications.

In this paper we adapt the nearest neighbour rule to the contextual bandit problem. Our algorithm handles the fully adversarial setting in which no assumptions at all are made about the data-generation process. When combined with a sufficiently fast data-structure for (perhaps approximate) adaptive nearest neighbour search, such as a navigating net, our algorithm is extremely efficient - having a per trial running time polylogarithmic in both the number of trials and actions, and taking only quasi-linear space. We give generic regret bounds for our algorithm and further analyse them when applied to the stochastic bandit problem in euclidean space. We note that our algorithm can also be applied to the online classification problem.

In this paper we consider the adversarial contextual bandit problem in metric spaces. The paper "Nearest neighbour with bandit feedback" tackled this problem but when there are many contexts near the decision boundary of the comparator policy it suffers from a high regret. In this paper we eradicate this problem, designing an algorithm in which we can hold out any set of contexts when computing our regret term. Our algorithm builds on that of "Nearest neighbour with bandit feedback" and hence inherits its extreme computational efficiency.

Communication networks able to withstand hostile environments are critically important for disaster relief operations. In this paper, we consider a challenging scenario where drones have been compromised in the supply chain, during their manufacture, and harbour malicious software capable of wide-ranging and infectious disruption. We investigate multi-agent deep reinforcement learning as a tool for learning defensive strategies that maximise communications bandwidth despite continual adversarial interference. Using a public challenge for learning network resilience strategies, we propose a state-of-the-art symbolic technique and study its superiority over deep reinforcement learning agents. Correspondingly, we identify three specific methods for improving the performance of our neural agents: (1) ensuring each observation contains the necessary information, (2) using symbolic agents to provide a curriculum for learning, and (3) paying close attention to reward. We apply our methods and present a new mixed strategy enabling symbolic and neural agents to work together and improve on all prior results.

As machine learning models become more capable, they have exhibited increased potential in solving complex tasks. One of the most promising directions uses deep reinforcement learning to train autonomous agents in computer network defense tasks. This work studies the impact of the reward signal that is provided to the agents when training for this task. Due to the nature of cybersecurity tasks, the reward signal is typically 1) in the form of penalties (e.g., when a compromise occurs), and 2) distributed sparsely across each defense episode. Such reward characteristics are atypical of classic reinforcement learning tasks where the agent is regularly rewarded for progress (cf. to getting occasionally penalized for failures). We investigate reward shaping techniques that could bridge this gap so as to enable agents to train more sample-efficiently and potentially converge to a better performance. We first show that deep reinforcement learning algorithms are sensitive to the magnitude of the penalties and their relative size. Then, we combine penalties with positive external rewards and study their effect compared to penalty-only training. Finally, we evaluate intrinsic curiosity as an internal positive reward mechanism and discuss why it might not be as advantageous for high-level network monitoring tasks.