Search for: All records

PowerShell-based malware increasingly leverages fileless, in-memory execution and advanced obfuscation techniques to bypass traditional signature-based security solutions. Although machine learning (ML) methods have shown promise in detecting such threats, their limited interpretability often hinders security analysts’ ability to understand and validate detection results. Recent advancements in large language models (LLMs) demonstrate their potential for both threat detection and the generation of interpretable explanations. This paper investigates the application of LLMs for detecting malicious PowerShell scripts and producing human-interpretable explanations for their classification decisions. In our approach, LLMs are prompted to classify PowerShell scripts, localize specific code segments responsible for the malicious behavior, and generate natural language explanations that articulate the reasoning behind these classifications. To assess the contribution of the identified segments to the script’s maliciousness, we perform an ablation study to measure changes in VirusTotal community scores before and after removing (masking) these segments. A significant reduction in score suggests that the code segments likely contribute to the script’s malicious behavior. Our results show that LLMs are capable of identifying and explaining malicious PowerShell scripts, although performance varies across different models. 

Logic rules allow analysis of complex relationships to be expressed easily, especially for transitive relations in critical applications. However, understanding and predicting the efficiency of different inference methods remain challenging, even for simplest rules given different kinds of input data. This paper analyzes the efficiency of all three types of well-known inference methods---query-driven, ground-and-solve, and fact-driven---along with their respective optimizations, and compares with optimal complexities for the first time, for analyzing transitive graph relations. We also experiment with rule systems widely considered to have the best performance. We analyze all well-known rule variants and widely varying input graphs. The results include precisely calculated optimal time complexities; comparative analysis across different inference methods, rule variants, and graph types; confirmation with performance experiments; as well as discovery of a performance bug. 

Advanced Persistent Threat (APT) attacks are sophisticated cyberattacks characterized by stealth, persistence, and long-term engagement with targeted systems. Traditional detection approaches using machine learning and deep learning rely on internal model representations or post hoc explainability techniques, which often lack human-readable context and require significant manual interpretation. This paper investigates the use of large language models (LLMs) for APT detection through code analysis. We evaluate LLMs’ ability to identify APT-related behaviors in code-snippet sequences, localize malicious components, and map them to corresponding MITRE ATT&CK tactics, techniques, and procedures (TTPs). To address context-window limitations, we introduce a segmentation-based approach that divides long code sequences into smaller segments and iteratively processes them while preserving context through summary propagation. Because real-world APT data are scarce due to their stealthy nature, privacy constraints, and long attack lifecycles, we construct an APT dataset derived from Caldera adversary profiles and emulation plans. Experimental results show that LLMs, when guided by rubric-based prompts and supplemented with ATT&CK domain knowledge, achieve robust performance across detection, localization, and TTP mapping tasks. 

Robust access control is a cornerstone of secure software, systems, and networks. An access control mechanism is as effective as the policy it enforces. However, authoring effective policies that satisfy desired properties such as the principle of least privilege is a challenging task even for experienced administrators. In this paper, we set out to address this pain point by proposing Restricter, which automatically tightens each (permit) policy rule of a policy with respect to an access log, which captures some already exercised access requests and their corresponding access decisions (i.e., allow or deny). Restricter achieves policy tightening by reducing the number of access requests permitted by a policy rule without sacrificing the functionality of the underlying system it is regulating. We implement Restricter for Amazon’s Cedar policy language and demonstrate its effectiveness through two realistic case studies. 

This paper presents a powerful automated framework for making complex systems resilient under failures, by optimized adaptive distribution and replication of interdependent software components across heterogeneous hardware components with widely varying capabilities. A configuration specifies how software is distributed and replicated: which software components to run on each computer, which software components to replicate, which replication protocols to use, etc. We present an algorithm that, given a system model and resilience requirements, (1) determines initial configurations of the system that are resilient, and (2) generates a reconfiguration policy that determines reconfiguration actions to execute in response to failures and recoveries. This model-finding algorithm is based on state-space exploration and incorporates powerful optimizations, including a quotient reduction based on a novel equivalence relation between states. We present experimental results from successfully applying a prototype implementation of our framework to a model of an autonomous driving system. 

In Proceedings of the 40th International Conference on Logic Programming (ICLP 2024). https://dx.doi.org/10.4204/EPTCS.416 

We present distributed distance-based control (DDC), a novel approach for controlling a multi-agent system, such that it achieves a desired formation, in a resource-constrained setting. Our controller is fully distributed and only requires local state-estimation and scalar measurements of inter-agent distances. It does not require an external localization system or inter-agent exchange of state information. Our approach uses spatial- predictive control (SPC), to optimize a cost function given strictly in terms of inter-agent distances and the distance to the target location. In DDC, each agent continuously learns and updates a very abstract model of the actual system, in the form of a dictionary of three independent key-value pairs (~s, d), where d is the partial derivative of the distance measurements along a spatial direction ~s. This is sufficient for an agent to choose the best next action. We validate our approach by using DDC to control a collection of Crazyflie drones to achieve formation flight and reach a target while maintaining flock formation. 

The Simplex Architecture is a runtime assurance framework where control authority may switch from an unverified and potentially unsafe advanced controller to a backup baseline controller in order to maintain the safety of an autonomous cyber-physical system. In this work, we show that runtime checks can replace the requirement to statically verify safety of the baseline controller. This is important as there are many powerful control techniques, such as model-predictive control and neural network controllers, that work well in practice but are difficult to statically verify. Since the method does not use internal information about the advanced or baseline controller, we call the approach the Black-Box Simplex Architecture. We prove the architecture is safe and present two case studies where (i) modelpredictive control provides safe multi-robot coordination, and (ii) neural networks provably prevent collisions in groups of F-16 aircraft, despite the controllers occasionally outputting unsafe commands. We further show how to safely blend commands from the advanced and baseline controllers in multiagent systems, reducing the performance impact when switching is necessary to preserve safety.