arXiv subjects

Surya Nepal

Publications and source records attributed to Surya Nepal.

At least 37 records · Page 2Linked to original sources

arXiv2025

Adversarial Attacks Against Automated Fact-Checking: A Survey

In an era where misinformation spreads freely, fact-checking (FC) plays a crucial role in verifying claims and promoting reliable information. While automated fact-checking (AFC) has advanced significantly, existing systems remain vulnerable to adversarial attacks that manipulate or generate claims, evidence, or claim-evidence pairs. These attacks can distort the truth, mislead decision-makers, and ultimately undermine the reliability of FC models. Despite growing research interest in adversarial attacks against AFC systems, a comprehensive, holistic overview of key challenges remains lacking. These challenges include understanding attack strategies, assessing the resilience of current models, and identifying ways to enhance robustness. This survey provides the first in-depth review of adversarial attacks targeting FC, categorizing existing attack methodologies and evaluating their impact on AFC systems. Additionally, we examine recent advancements in adversary-aware defenses and highlight open research questions that require further exploration. Our findings underscore the urgent need for resilient FC frameworks capable of withstanding adversarial manipulations in pursuit of preserving high verification accuracy.

cs.CL↗

arXiv2025

RINSER: Accurate API Prediction Using Masked Language Models

Malware authors commonly use obfuscation to hide API identities in binary files, making analysis difficult and time-consuming for a human expert to understand the behavior and intent of the program. Automatic API prediction tools are necessary to efficiently analyze unknown binaries, facilitating rapid malware triage while reducing the workload on human analysts. In this paper, we present RINSER (AccuRate API predictioN using maSked languagE model leaRning), an automated framework for predicting Windows API (WinAPI) function names. RINSER introduces the novel concept of API codeprints, a set of API-relevant assembly instructions, and supports x86 PE binaries. RINSER relies on BERT's masked language model (LM) to predict API names at scale, achieving 85.77% accuracy for normal binaries and 82.88% accuracy for stripped binaries. We evaluate RINSER on a large dataset of 4.7M API codeprints from 11,098 malware binaries, covering 4,123 unique Windows APIs, making it the largest publicly available dataset of this type. RINSER successfully discovered 65 obfuscated Windows APIs related to C2 communication, spying, and evasion in our dataset, which the commercial disassembler IDA failed to identify. Furthermore, we compared RINSER against three state-of-the-art approaches, showing over 20% higher prediction accuracy. We also demonstrated RINSER's resilience to adversarial attacks, including instruction randomization and code displacement, with a performance drop of no more than 3%.

cs.CY↗

arXiv2025

Towards Faithful Class-level Self-explainability in Graph Neural Networks by Subgraph Dependencies

Enhancing the interpretability of graph neural networks (GNNs) is crucial to ensure their safe and fair deployment. Recent work has introduced self-explainable GNNs that generate explanations as part of training, improving both faithfulness and efficiency. Some of these models, such as ProtGNN and PGIB, learn class-specific prototypes, offering a potential pathway toward class-level explanations. However, their evaluations focus solely on instance-level explanations, leaving open the question of whether these prototypes meaningfully generalize across instances of the same class. In this paper, we introduce GraphOracle, a novel self-explainable GNN framework designed to generate and evaluate class-level explanations for GNNs. Our model jointly learns a GNN classifier and a set of structured, sparse subgraphs that are discriminative for each class. We propose a novel integrated training that captures graph$\unicode{x2013}$subgraph$\unicode{x2013}$prediction dependencies efficiently and faithfully, validated through a masking-based evaluation strategy. This strategy enables us to retroactively assess whether prior methods like ProtGNN and PGIB deliver effective class-level explanations. Our results show that they do not. In contrast, GraphOracle achieves superior fidelity, explainability, and scalability across a range of graph classification tasks. We further demonstrate that GraphOracle avoids the computational bottlenecks of previous methods$\unicode{x2014}$like Monte Carlo Tree Search$\unicode{x2014}$by using entropy-regularized subgraph selection and lightweight random walk extraction, enabling faster and more scalable training. These findings position GraphOracle as a practical and principled solution for faithful class-level self-explainability in GNNs.

cs.LG↗

arXiv2025

Adaptive alert prioritisation in security operations centres via learning to defer with human feedback

Alert prioritisation (AP) is crucial for security operations centres (SOCs) to manage the overwhelming volume of alerts and ensure timely detection and response to genuine threats, while minimising alert fatigue. Although predictive AI can process large alert volumes and identify known patterns, it struggles with novel and evolving scenarios that demand contextual understanding and nuanced judgement. A promising solution is Human-AI teaming (HAT), which combines human expertise with AI's computational capabilities. Learning to Defer (L2D) operationalises HAT by enabling AI to "defer" uncertain or unfamiliar cases to human experts. However, traditional L2D models rely on static deferral policies that do not evolve with experience, limiting their ability to learn from human feedback and adapt over time. To overcome this, we introduce Learning to Defer with Human Feedback (L2DHF), an adaptive deferral framework that leverages Deep Reinforcement Learning from Human Feedback (DRLHF) to optimise deferral decisions. By dynamically incorporating human feedback, L2DHF continuously improves AP accuracy and reduces unnecessary deferrals, enhancing SOC effectiveness and easing analyst workload. Experiments on two widely used benchmark datasets, UNSW-NB15 and CICIDS2017, demonstrate that L2DHF significantly outperforms baseline models. Notably, it achieves 13-16% higher AP accuracy for critical alerts on UNSW-NB15 and 60-67% on CICIDS2017. It also reduces misprioritisations, for example, by 98% for high-category alerts on CICIDS2017. Moreover, L2DHF decreases deferrals, for example, by 37% on UNSW-NB15, directly reducing analyst workload. These gains are achieved with efficient execution, underscoring L2DHF's practicality for real-world SOC deployment.

cs.CR↗

arXiv2025

What is the Cost of Differential Privacy for Deep Learning-Based Trajectory Generation?

While location trajectories offer valuable insights, they also reveal sensitive personal information. Differential Privacy (DP) offers formal protection, but achieving a favourable utility-privacy trade-off remains challenging. Recent works explore deep learning-based generative models to produce synthetic trajectories. However, current models lack formal privacy guarantees and rely on conditional information derived from real data during generation. This work investigates the utility cost of enforcing DP in such models, addressing three research questions across two datasets and eleven utility metrics. (1) We evaluate how DP-SGD, the standard DP training method for deep learning, affects the utility of state-of-the-art generative models. (2) Since DP-SGD is limited to unconditional models, we propose a novel DP mechanism for conditional generation that provides formal guarantees and assess its impact on utility. (3) We analyse how model types - Diffusion, VAE, and GAN - affect the utility-privacy trade-off. Our results show that DP-SGD significantly impacts performance, although some utility remains if the datasets is sufficiently large. The proposed DP mechanism improves training stability, particularly when combined with DP-SGD, for unstable models such as GANs and on smaller datasets. Diffusion models yield the best utility without guarantees, but with DP-SGD, GANs perform best, indicating that the best non-private model is not necessarily optimal when targeting formal guarantees. In conclusion, DP trajectory generation remains a challenging task, and formal guarantees are currently only feasible with large datasets and in constrained use cases.

cs.CR↗

arXiv2025

ContextBuddy: AI-Enhanced Contextual Insights for Security Alert Investigation (Applied to Intrusion Detection)

Modern Security Operations Centres (SOCs) integrate diverse tools, such as SIEM, IDS, and XDR systems, offering rich contextual data, including alert enrichments, flow features, and similar case histories. Yet, analysts must still manually determine which of these contextual cues are most relevant when validating specific alerts. We introduce ContextBuddy, an AI assistant that learns from analysts' prior investigations to help them identify the most relevant context for new alerts. Rather than providing enrichments, ContextBuddy models how analysts have previously selected context and suggests tailored cues based on the characteristics of each alert. We formulate context selection as a sequential decision-making problem and apply imitation learning (IL) to capture analysts' strategies, evaluating multiple IL approaches. Through staged evaluation, we validate ContextBuddy using two intrusion detection datasets (HIKARI-2021, UNSW-NB15). In simulation-based experiments, ContextBuddy helped simulated reinforcement learning analysts improve classification accuracy (p < 0.001) (increasing F1 by 2.5% for HIKARI and 9% for UNSW), reducing false negatives (1.5% for HIKARI and 10% for UNSW), and keeping false positives below 1%. Decision confidence among agents also improved by 2-3% (p < 0.001). In a within-subject user study (N=13; power = 0.8), non-experts using ContextBuddy improved classification accuracy by 21.1% (p = 0.008) and reduced alert validation time by 24% (p = 0.01). These results demonstrate that by learning context-selection patterns from analysts, ContextBuddy can yield notable improvements in investigation effectiveness and efficiency.

cs.CR↗

arXiv2025

Bridging Expertise Gaps: The Role of LLMs in Human-AI Collaboration for Cybersecurity

This study investigates whether large language models (LLMs) can function as intelligent collaborators to bridge expertise gaps in cybersecurity decision-making. We examine two representative tasks-phishing email detection and intrusion detection-that differ in data modality, cognitive complexity, and user familiarity. Through a controlled mixed-methods user study, n = 58 (phishing, n = 34; intrusion, n = 24), we find that human-AI collaboration improves task performance,reducing false positives in phishing detection and false negatives in intrusion detection. A learning effect is also observed when participants transition from collaboration to independent work, suggesting that LLMs can support long-term skill development. Our qualitative analysis shows that interaction dynamics-such as LLM definitiveness, explanation style, and tone-influence user trust, prompting strategies, and decision revision. Users engaged in more analytic questioning and showed greater reliance on LLM feedback in high-complexity settings. These results provide design guidance for building interpretable, adaptive, and trustworthy human-AI teaming systems, and demonstrate that LLMs can meaningfully support non-experts in reasoning through complex cybersecurity problems.

cs.CR↗

arXiv2025

Saving Storage Space Using Files on the Web

As conventional storage density reaches its physical limits, the cost of a gigabyte of storage is no longer plummeting, but rather has remained mostly flat for the past decade. Meanwhile, file sizes continue to grow, leading to ever fuller drives. When a user's storage is full, they must disrupt their workflow to laboriously find large files that are good candidates for deletion. Separately, the web acts as a distributed storage network, providing free access to petabytes of redundant files across 200 million websites. An automated method of restoring files from the web would enable more efficient storage management, since files readily recoverable from the web would make good candidates for removal. Despite this, there are no prescribed methods for automatically detecting these files and ensuring their easy recoverability from the web, as little is known about either the biggest files of users or their origins on the web. This study thus seeks to determine what files consume the most space in users' storage, and from this, to propose an automated method to select candidate files for removal. Our investigations show 989 MB of storage per user can be saved by inspecting preexisting metadata of their 25 largest files alone, with file recovery from the web 3 months later. This demonstrates the feasibility of applying such a method in a climate of increasingly scarce local storage resources.

cs.OS↗

arXiv2025

Enhancing Malware Fingerprinting through Analysis of Evasive Techniques

As malware detection evolves, attackers adopt sophisticated evasion tactics. Traditional file-level fingerprinting, such as cryptographic and fuzzy hashes, is often overlooked as a target for evasion. Malware variants exploit minor binary modifications to bypass detection, as seen in Microsoft's discovery of GoldMax variations (2020-2021). However, no large-scale empirical studies have assessed the limitations of traditional fingerprinting methods on real-world malware samples or explored improvements. This paper fills this gap by addressing three key questions: (a) How prevalent are file variants in malware samples? Analyzing 4 million Windows Portable Executable (PE) files, 21 million sections, and 48 million resources, we find up to 80% deep structural similarities, including common APIs and executable sections. (b) What evasion techniques are used? We identify resilient fingerprints (clusters of malware variants with high similarity) validated via VirusTotal. Our analysis reveals non-functional mutations, such as altered section numbers, virtual sizes, and section names, as primary evasion tactics. We also classify two key section types: malicious sections (high entropy >5) and camouflage sections (entropy = 0). (c) How can fingerprinting be improved? We propose two novel approaches that enhance detection, improving identification rates from 20% (traditional methods) to over 50% using our refined fingerprinting techniques. Our findings highlight the limitations of existing methods and propose new strategies to strengthen malware fingerprinting against evolving threats.

cs.CR↗

arXiv2025

Provisioning Time-Based Subscription in NDN: A Secure and Efficient Access Control Scheme

This paper proposes a novel encryption-based access control mechanism for Named Data Networking (NDN). The scheme allows data producers to share their content in encrypted form before transmitting it to consumers. The encryption mechanism incorporates time-based subscription access policies directly into the encrypted content, enabling only consumers with valid subscriptions to decrypt it. This makes the scheme well-suited for real-world, subscription-based applications like Netflix. Additionally, the scheme introduces an anonymous and unlinkable signature-based authentication mechanism that empowers edge routers to block bogus content requests at the network's entry point, thereby mitigating Denial of Service (DoS) attacks. A formal security proof demonstrates the scheme's resistance to Chosen Plaintext Attacks (CPA). Performance analysis, using Mini-NDN-based emulation and a Charm library implementation, further confirms the practicality of the scheme. Moreover, it outperforms closely related works in terms of functionality, security, and communication overhead.

cs.CR↗

arXiv2025

Comprehensive Evaluation of Cloaking Backdoor Attacks on Object Detector in Real-World

The exploration of backdoor vulnerabilities in object detectors, particularly in real-world scenarios, remains limited. A significant challenge lies in the absence of a natural physical backdoor dataset, and constructing such a dataset is both time- and labor-intensive. In this work, we address this gap by creating a large-scale dataset comprising approximately 11,800 images/frames with annotations featuring natural objects (e.g., T-shirts and hats) as triggers to incur cloaking adversarial effects in diverse real-world scenarios. This dataset is tailored for the study of physical backdoors in object detectors. Leveraging this dataset, we conduct a comprehensive evaluation of an insidious cloaking backdoor effect against object detectors, wherein the bounding box around a person vanishes when the individual is near a natural object (e.g., a commonly available T-shirt) in front of the detector. Our evaluations encompass three prevalent attack surfaces: data outsourcing, model outsourcing, and the use of pretrained models. The cloaking effect is successfully implanted in object detectors across all three attack surfaces. We extensively evaluate four popular object detection algorithms (anchor-based Yolo-V3, Yolo-V4, Faster R-CNN, and anchor-free CenterNet) using 19 videos (totaling approximately 11,800 frames) in real-world scenarios. Our results demonstrate that the backdoor attack exhibits remarkable robustness against various factors, including movement, distance, angle, non-rigid deformation, and lighting. In data and model outsourcing scenarios, the attack success rate (ASR) in most videos reaches 100% or near it, while the clean data accuracy of the backdoored model remains indistinguishable from that of the clean model, making it impossible to detect backdoor behavior through a validation set.

cs.CR↗

arXiv2025

OblivCDN: A Practical Privacy-preserving CDN with Oblivious Content Access

Content providers increasingly utilise Content Delivery Networks (CDNs) to enhance users' content download experience. However, this deployment scenario raises significant security concerns regarding content confidentiality and user privacy due to the involvement of third-party providers. Prior proposals using private information retrieval (PIR) and oblivious RAM (ORAM) have proven impractical due to high computation and communication costs, as well as integration challenges within distributed CDN architectures. In response, we present \textsf{OblivCDN}, a practical privacy-preserving system meticulously designed for seamless integration with the existing real-world Internet-CDN infrastructure. Our design strategically adapts Range ORAM primitives to optimise memory and disk seeks when accessing contiguous blocks of CDN content, both at the origin and edge servers, while preserving both content confidentiality and user access pattern hiding features. Also, we carefully customise several oblivious building blocks that integrate the distributed trust model into the ORAM client, thereby eliminating the computational bottleneck in the origin server and reducing communication costs between the origin server and edge servers. Moreover, the newly-designed ORAM client also eliminates the need for trusted hardware on edge servers, and thus significantly ameliorates the compatibility towards networks with massive legacy devices.In real-world streaming evaluations, OblivCDN} demonstrates remarkable performance, downloading a $256$ MB video in just $5.6$ seconds. This achievement represents a speedup of $90\times$ compared to a strawman approach (direct ORAM adoption) and a $366\times$ improvement over the prior art, OblivP2P.

cs.CR↗

arXiv2024

Analysing one- and two-bit data to reduce memory requirements for F-statistic-based gravitational wave searches

Searches for continuous-wave gravitational radiation in data collected by modern long-baseline interferometers, such as the Laser Interferometer Gravitational-wave Observatory (LIGO), the Virgo interferometer and the Kamioka Gravitational Wave Detector (KAGRA), can be memory intensive. A digitisation scheme is described that reduces the 64-bit interferometer output to a one- or two-bit data stream while minimising distortion and achieving considerable reduction in storage and input/output cost. For the representative example of the coherent, maximum-likelihood matched filter known as the F-statistic, it is found using Monte-Carlo simulations that the injected signal only needs to be ~24 per cent stronger (for one-bit data) and ~6.4 per cent stronger (for two bit data with optimal thresholds) than a 64-bit signal in order to be detected with 90 per cent probability in Gaussian noise. The foregoing percentages do not change significantly when the signal frequency decreases secularly, or when the noise statistics are not Gaussian, as verified with LIGO Science Run 6 data.

gr-qc↗

arXiv2024

From Solitary Directives to Interactive Encouragement! LLM Secure Code Generation by Natural Language Prompting

Large Language Models (LLMs) have shown remarkable potential in code generation, making them increasingly important in the field. However, the security issues of generated code have not been fully addressed, and the usability of LLMs in code generation still requires further exploration. This work introduces SecCode, a framework that leverages an innovative interactive encouragement prompting (EP) technique for secure code generation with \textit{only NL} prompts. This approach ensures that the prompts can be easily shared and understood by general users. SecCode functions through three stages: 1) Code Generation using NL Prompts; 2) Code Vulnerability Detection and Fixing, utilising our proposed encouragement prompting; 3) Vulnerability Cross-Checking and Code Security Refinement. These stages are executed in multiple interactive iterations to progressively enhance security. By using both proprietary LLMs (i.e., GPT-3.5 Turbo, GPT-4 and GPT-4o) and open-source LLMs (i.e., Llama 3.1 8B Instruct, DeepSeek Coder V2 Lite Instruct) evaluated on three benchmark datasets, extensive experimental results show that our proposed SecCode greatly outperforms compared baselines, generating secure code with a high vulnerability correction rate. For example, SecCode exhibits a high fix success rate of over 76\% after running 5 automated EP interactive iterations and over 89\% after running 10 automated EP interactive iterations. To the best of our knowledge, this work is the first to formulate secure code generation with NL prompts only. We have open-sourced our code and encourage the community to focus on secure code generation.

cs.CR↗

arXiv2024

Adversarially Guided Stateful Defense Against Backdoor Attacks in Federated Deep Learning

Recent works have shown that Federated Learning (FL) is vulnerable to backdoor attacks. Existing defenses cluster submitted updates from clients and select the best cluster for aggregation. However, they often rely on unrealistic assumptions regarding client submissions and sampled clients population while choosing the best cluster. We show that in realistic FL settings, state-of-the-art (SOTA) defenses struggle to perform well against backdoor attacks in FL. To address this, we highlight that backdoored submissions are adversarially biased and overconfident compared to clean submissions. We, therefore, propose an Adversarially Guided Stateful Defense (AGSD) against backdoor attacks on Deep Neural Networks (DNNs) in FL scenarios. AGSD employs adversarial perturbations to a small held-out dataset to compute a novel metric, called the trust index, that guides the cluster selection without relying on any unrealistic assumptions regarding client submissions. Moreover, AGSD maintains a trust state history of each client that adaptively penalizes backdoored clients and rewards clean clients. In realistic FL settings, where SOTA defenses mostly fail to resist attacks, AGSD mostly outperforms all SOTA defenses with minimal drop in clean accuracy (5% in the worst-case compared to best accuracy) even when (a) given a very small held-out dataset -- typically AGSD assumes 50 samples (<= 0.1% of the training data) and (b) no heldout dataset is available, and out-of-distribution data is used instead. For reproducibility, our code will be openly available at: https://github.com/hassanalikhatim/AGSD.

cs.LG↗

arXiv2024

SAFE: Advancing Large Language Models in Leveraging Semantic and Syntactic Relationships for Software Vulnerability Detection

Software vulnerabilities (SVs) have emerged as a prevalent and critical concern for safety-critical security systems. This has spurred significant advancements in utilizing AI-based methods, including machine learning and deep learning, for software vulnerability detection (SVD). While AI-based methods have shown promising performance in SVD, their effectiveness on real-world, complex, and diverse source code datasets remains limited in practice. To tackle this challenge, in this paper, we propose a novel framework that enhances the capability of large language models to learn and utilize semantic and syntactic relationships from source code data for SVD. As a result, our approach can enable the acquisition of fundamental knowledge from source code data while adeptly utilizing crucial relationships, i.e., semantic and syntactic associations, to effectively address the software vulnerability detection (SVD) problem. The rigorous and extensive experimental results on three real-world challenging datasets (i.e., ReVeal, D2A, and Devign) demonstrate the superiority of our approach over the effective and state-of-the-art baselines. In summary, on average, our SAFE approach achieves higher performances from 4.79% to 9.15% for F1-measure and from 16.93% to 21.70% for Recall compared to the baselines across all datasets used.

cs.SE↗

arXiv2024

SoK: Can Trajectory Generation Combine Privacy and Utility?

While location trajectories represent a valuable data source for analyses and location-based services, they can reveal sensitive information, such as political and religious preferences. Differentially private publication mechanisms have been proposed to allow for analyses under rigorous privacy guarantees. However, the traditional protection schemes suffer from a limiting privacy-utility trade-off and are vulnerable to correlation and reconstruction attacks. Synthetic trajectory data generation and release represent a promising alternative to protection algorithms. While initial proposals achieve remarkable utility, they fail to provide rigorous privacy guarantees. This paper proposes a framework for designing a privacy-preserving trajectory publication approach by defining five design goals, particularly stressing the importance of choosing an appropriate Unit of Privacy. Based on this framework, we briefly discuss the existing trajectory protection approaches, emphasising their shortcomings. This work focuses on the systematisation of the state-of-the-art generative models for trajectories in the context of the proposed framework. We find that no existing solution satisfies all requirements. Thus, we perform an experimental study evaluating the applicability of six sequential generative models to the trajectory domain. Finally, we conclude that a generative trajectory model providing semantic guarantees remains an open research question and propose concrete next steps for future research.

cs.CR↗

arXiv2024

Watch Out! Simple Horizontal Class Backdoor Can Trivially Evade Defense

All current backdoor attacks on deep learning (DL) models fall under the category of a vertical class backdoor (VCB) -- class-dependent. In VCB attacks, any sample from a class activates the implanted backdoor when the secret trigger is present. Existing defense strategies overwhelmingly focus on countering VCB attacks, especially those that are source-class-agnostic. This narrow focus neglects the potential threat of other simpler yet general backdoor types, leading to false security implications. This study introduces a new, simple, and general type of backdoor attack coined as the horizontal class backdoor (HCB) that trivially breaches the class dependence characteristic of the VCB, bringing a fresh perspective to the community. HCB is now activated when the trigger is presented together with an innocuous feature, regardless of class. For example, the facial recognition model misclassifies a person who wears sunglasses with a smiling innocuous feature into the targeted person, such as an administrator, regardless of which person. The key is that these innocuous features are horizontally shared among classes but are only exhibited by partial samples per class. Extensive experiments on attacking performance across various tasks, including MNIST, facial recognition, traffic sign recognition, object detection, and medical diagnosis, confirm the high efficiency and effectiveness of the HCB. We rigorously evaluated the evasiveness of the HCB against a series of eleven representative countermeasures, including Fine-Pruning (RAID 18'), STRIP (ACSAC 19'), Neural Cleanse (Oakland 19'), ABS (CCS 19'), Februus (ACSAC 20'), NAD (ICLR 21'), MNTD (Oakland 21'), SCAn (USENIX SEC 21'), MOTH (Oakland 22'), Beatrix (NDSS 23'), and MM-BD (Oakland 24'). None of these countermeasures prove robustness, even when employing a simplistic trigger, such as a small and static white-square patch.

cs.CR↗