Papers

Large Language Models (LLMs) are often fine-tuned to adapt their general-purpose knowledge to specific tasks and domains such as cyber threat intelligence (CTI). Fine-tuning is mostly done through proprietary datasets that may contain sensitive information. Owners expect their fine-tuned model to not inadvertently leak this information to potentially adversarial end users. Using CTI as a use case, we demonstrate that data-extraction attacks can recover sensitive information from fine-tuned models on CTI reports, underscoring the need for mitigation. Retraining the full model to eliminate this leakage is computationally expensive and impractical. We propose an alternative approach, which we call privacy alignment, inspired by safety alignment in LLMs. Just like safety alignment teaches the model to abide by safety constraints through a few examples, we enforce privacy alignment through few-shot supervision, integrating a privacy classifier and a privacy redactor, both handled by the same underlying LLM. We evaluate our system, called CTIGuardian, using GPT-4o mini and Mistral-7B Instruct models, benchmarking against Presidio, a named entity recognition (NER) baseline. Results show that CTIGuardian provides a better privacy-utility trade-off than NER based models. While we demonstrate its effectiveness on a CTI use case, the framework is generic enough to be applicable to other sensitive domains.

As post-quantum cryptography (PQC) becomes increasingly critical for securing future communication systems, the performance overhead introduced by quantum-resistant algorithms presents a major computing challenge. HQC (Hamming Quasi-Cyclic) is a newly standardized code-based PQC scheme designed to replace classical key exchange methods. In this paper, we propose OptHQC, an optimized implementation of the HQC scheme to deliver high-performance cryptographic operations. Our approach provides a comprehensive analysis of each computational blocks in HQC and introduces optimizations across all three stages: key generation, encryption, and decryption. We first exploit data-level sparsity in vector multiplication to accelerate polynomial operations during vector generation. We then leverage instruction-level acceleration (e.g., AVX2) in hash computation to further improve performance. Last, we transform multiplication into lookup table indexing and optimize memory access patterns in syndrome computation and error vector recovery, which are the most computationally intensive operations in HQC. Overall, OptHQC achieves an average 55% speedup over the reference HQC implementation on CPU.

Alpha-Root is a cybersecurity-focused dataset collected in a single shot from the Common Crawl web graph using community detection. Unlike iterative content-scoring approaches like DeepSeekMath, we mine quality domains directly from the web graph, starting from just 20 trusted seed domains.

Large language models (LLMs) are increasingly deployed behind safety guardrails such as system prompts and content filters, especially in settings where product teams cannot modify model weights. In practice these guardrails are typically hand-tuned, brittle, and difficult to reproduce. This paper studies a simple but practical alternative: treat safety guardrail design itself as a hyperparameter optimization problem over a frozen base model. Concretely, I wrap Mistral-7B-Instruct with modular jailbreak and malware system prompts plus a ModernBERT-based harmfulness classifier, then evaluate candidate configurations on three public benchmarks covering malware generation, classic jailbreak prompts, and benign user queries. Each configuration is scored using malware and jailbreak attack success rate, benign harmful-response rate, and end-to-end latency. A 48-point grid search over prompt combinations and filter modes establishes a baseline. I then run a black-box Optuna study over the same space and show that it reliably rediscovers the best grid configurations while requiring an order of magnitude fewer evaluations and roughly 8x less wall-clock time. The results suggest that viewing safety guardrails as tunable hyperparameters is a feasible way to harden black-box LLM deployments under compute and time constraints.

This project presents a unified detection framework that constructs a complete corpus of Microsoft Graph permissions, generates consistent LLM-based risk scores, and integrates them into a real-time detection engine to identify malicious OAuth consent activity.

We evaluate adversarial robustness in tabular machine learning models used in financial decision making. Using credit scoring and fraud detection data, we apply gradient based attacks and measure impacts on discrimination, calibration, and financial risk metrics. Results show notable performance degradation under small perturbations and partial recovery through adversarial training.

Network Intrusion Detection Systems (NIDS) are essential for securing networks by identifying and mitigating unauthorized activities indicative of cyberattacks. As cyber threats grow increasingly sophisticated, NIDS must evolve to detect both emerging threats and deviations from normal behavior. This study explores the application of machine learning (ML) methods to improve the NIDS accuracy through analyzing intricate structures in deep-featured network traffic records. Leveraging the 1999 KDD CUP intrusion dataset as a benchmark, this research evaluates and optimizes several ML algorithms, including Support Vector Machines (SVM), Naïve Bayes variants (MNB, BNB), Random Forest (RF), k-Nearest Neighbors (k-NN), Decision Trees (DT), AdaBoost, XGBoost, Logistic Regression (LR), Ridge Classifier, Passive-Aggressive (PA) Classifier, Rocchio Classifier, Artificial Neural Networks (ANN), and Perceptron (PPN). Initial evaluations without hyper-parameter optimization demonstrated suboptimal performance, highlighting the importance of tuning to enhance classification accuracy. After hyper-parameter optimization using grid and random search techniques, the SVM classifier achieved 99.12% accuracy with a 0.0091 False Alarm Rate (FAR), outperforming its default configuration (98.08% accuracy, 0.0123 FAR) and all other classifiers. This result confirms that SVM accomplishes the highest accuracy among the evaluated classifiers. We validated the effectiveness of all classifiers using a tenfold cross-validation approach, incorporating Recursive Feature Elimination (RFE) for feature selection to enhance the classifiers accuracy and efficiency. Our outcomes indicate that ML classifiers are both adaptable and reliable, contributing to enhanced accuracy in systems for detecting network intrusions.

State-space models (SSMs), exemplified by the Mamba architecture, have recently emerged as state-of-the-art sequence-modeling frameworks, offering linear-time scalability together with strong performance in long-context settings. Owing to their unique combination of efficiency, scalability, and expressive capacity, SSMs have become compelling alternatives to transformer-based models, which suffer from the quadratic computational and memory costs of attention mechanisms. As SSMs are increasingly deployed in real-world applications, it is critical to assess their susceptibility to both software- and hardware-level threats to ensure secure and reliable operation. Among such threats, hardware-induced bit-flip attacks (BFAs) pose a particularly severe risk by corrupting model parameters through memory faults, thereby undermining model accuracy and functional integrity. To investigate this vulnerability, we introduce RAMBO, the first BFA framework specifically designed to target Mamba-based architectures. Through experiments on the Mamba-1.4b model with LAMBADA benchmark, a cloze-style word-prediction task, we demonstrate that flipping merely a single critical bit can catastrophically reduce accuracy from 74.64% to 0% and increase perplexity from 18.94 to 3.75 x 10^6. These results demonstrate the pronounced fragility of SSMs to adversarial perturbations.

Modern computing systems face security threats, including memory corruption attacks, speculative execution vulnerabilities, and control-flow hijacking. Although existing solutions address these threats individually, they frequently introduce performance overhead and leave security gaps. This paper presents a Variable Record Table (VRT) with a unified hardware-assisted framework that simultaneously enforces spatial memory safety against buffer overflows, back-edge control-flow integrity (CFI), and speculative execution attack detection. The VRT dynamically constructs a protection table by instrumenting run-time instructions to extract memory addresses, bounds metadata, and control-flow signatures. Our evaluation across MiBench and SPEC benchmarks shows that VRT successfully detects all attack variants tested with zero additional instruction overhead. Furthermore, it maintains memory requirements below 25KB (for 512 entries) and maintains area / power overhead under 8% and 11.65 {\mu}W, respectively. By consolidating three essential security mechanisms into a single hardware structure, VRT provides comprehensive protection while minimizing performance impact.

Watermarking large language models (LLMs) is vital for preventing their misuse, including the fabrication of fake news, plagiarism, and spam. It is especially important to watermark LLM-generated code, as it often contains intellectual property.However, we found that existing methods for watermarking LLM-generated code fail to address comment removal attack.In such cases, an attacker can simply remove the comments from the generated code without affecting its functionality, significantly reducing the effectiveness of current code-watermarking techniques.On the other hand, injecting a watermark into code is challenging because, as previous works have noted, most code represents a low-entropy scenario compared to natural language. Our approach to addressing this issue involves leveraging prior knowledge to distinguish between low-entropy and high-entropy parts of the code, as indicated by a Cue List of words.We then inject the watermark guided by this Cue List, achieving higher detectability and usability than existing methods.We evaluated our proposed method on HumanEvaland compared our method with three state-of-the-art code watermarking techniques. The results demonstrate the effectiveness of our approach.

Finetuning pretrained large language models (LLMs) has become the standard paradigm for developing downstream applications. However, its security implications remain unclear, particularly regarding whether finetuned LLMs inherit jailbreak vulnerabilities from their pretrained sources. We investigate this question in a realistic pretrain-to-finetune threat model, where an attacker has full access to a released pretrained LLM but no access to its proprietary finetuned derivatives. Empirical analysis shows that adversarial prompts optimized on the pretrained model transfer most effectively to its finetuned variants, revealing inherited vulnerabilities from pretrained to finetuned LLMs. To further examine this inheritance, we conduct representation-level probing, which shows that transferable prompts are linearly separable within the pretrained hidden states, suggesting that transferability-relevant structure is already encoded in pretrained representations. Building on this insight, we propose the Probe-Guided Projection (PGP) attack, which steers optimization toward transferability-relevant directions. Experiments across multiple LLM families and diverse finetuned tasks confirm PGP's strong transfer success, underscoring the security risks inherent in the pretrain-to-finetune paradigm. Finally, we demonstrate that the same representation-level insights also enable a lightweight defense that mitigates pretrain-to-finetune jailbreak transfer while preserving downstream utility.

AI-powered greenwashing has emerged as an insidious challenge within corporate sustainability governance, exacerbating the opacity of environmental disclosures and subverting regulatory oversight. This study conducts a comparative legal analysis of criminal liability for AI-mediated greenwashing across India, the US, and the EU, exposing doctrinal lacunae in attributing culpability when deceptive claims originate from algorithmic systems. Existing statutes exhibit anthropocentric biases by predicating liability on demonstrable human intent, rendering them ill-equipped to address algorithmic deception. The research identifies a critical gap in jurisprudential adaptation, as prevailing fraud statutes remain antiquated vis-à-vis AI-generated misrepresentation. Utilising a doctrinal legal methodology, this study systematically dissects judicial precedents and statutory instruments, yielding results regarding the potential expansion of corporate criminal liability. Findings underscore the viability of strict liability models, recalibrated governance frameworks for AI accountability, and algorithmic due diligence mandates under ESG regimes. Comparative insights reveal jurisdictional disparities, with the EU Corporate Sustainability Due Diligence Directive (CSDDD) offering a potential transnational model. This study contributes to AI ethics and environmental jurisprudence by advocating for a hybrid liability framework integrating algorithmic risk assessment with legal personhood constructs, ensuring algorithmic opacity does not preclude liability enforcement.

Space infrastructures represent an emerging domain that is critical to the global economy and society. However, this domain is vulnerable to attacks, including cyber attacks and other kinds of attacks. To enhance the resilience of this domain, we must understand these attacks that can be waged against it and the defenses that can be employed to mitigate these attacks. The status quo is that there is neither a systematic understanding of these attacks against, nor defenses for, space infrastructures, despite their clear importance in guiding systematic analysis of space security and future research. In this paper, we fill the void by proposing the first systematic taxonomy of attacks against, and defenses for, space infrastructures. We hope this paper will inspire a community effort at refining the taxonomy towards a widely used one.

Despite their remarkable performance, deep neural networks exhibit a critical vulnerability: small, often imperceptible, adversarial perturbations can lead to drastically altered model predictions. Given the stringent reliability demands of applications such as medical diagnosis and autonomous driving, robust detection of such adversarial attacks is paramount. In this paper, we investigate the geometric properties of a model's input loss landscape. We analyze the Intrinsic Dimensionality (ID) of the model's gradient parameters, which quantifies the minimal number of coordinates required to describe the data points on their underlying manifold. We reveal a distinct and consistent difference in the ID for natural and adversarial data, which forms the basis of our proposed detection method. We validate our approach across two distinct operational scenarios. First, in a batch-wise context for identifying malicious data groups, our method demonstrates high efficacy on datasets like MNIST and SVHN. Second, in the critical individual-sample setting, we establish new state-of-the-art results on challenging benchmarks such as CIFAR-10 and MS COCO. Our detector significantly surpasses existing methods against a wide array of attacks, including CW and AutoAttack, achieving detection rates consistently above 92% on CIFAR-10. The results underscore the robustness of our geometric approach, highlighting that intrinsic dimensionality is a powerful fingerprint for adversarial detection across diverse datasets and attack strategies.

The widespread adoption of embedded systems has led to their deployment in critical real-world applications, making them attractive targets for malicious actors. These devices face unique challenges in mitigating vulnerabilities due to intrinsic constraints, such as low energy consumption requirements and limited computational resources. This paper presents RunPBA, a hardware-based runtime attestation system designed to defend against control flow attacks while maintaining minimal performance overhead and adhering to strict power consumption constraints. RunPBA leverages PACBTI, a new processor extension tailored for the Arm Cortex M processor family, allowing robust protection without requiring hardware modifications, a limitation present in similar solutions. We implemented a proof-of-concept and evaluated it using two benchmark suites. Experimental results indicate that RunPBA imposes a geometric mean performance overhead of only 1% and 4.7% across the benchmarks, underscoring its efficiency and suitability for real-world deployment.

Browser-using agents (BUAs) are an emerging class of AI agents that interact with web browsers in human-like ways, including clicking, scrolling, filling forms, and navigating across pages. While these agents help automate repetitive online tasks, they are vulnerable to prompt injection attacks that trick an agent into performing undesired actions, such as leaking private information or issuing unintended state-changing requests. We propose ceLLMate, a browser-level sandboxing framework that restricts the agent's ambient authority and reduces the blast radius of prompt injections. We address the semantic gap challenge that is fundamental to BUAs – writing and enforcing security policies for low-level UI tools like clicks and keystrokes is brittle and error-prone. Our core insight is to perform sandboxing at the HTTP layer because all side-effecting UI operations will result in network communication to the website's backend. We implement ceLLMate as an agent-agnostic browser extension and demonstrate how it enables sandboxing policies that block prompt injection attacks in the WASP benchmark with 7.25–15% latency overhead.

The increasing reliance on software in various applications has made the problem of software vulnerability detection more critical. Software vulnerabilities can lead to security breaches, data theft, and other negative outcomes. Traditional software vulnerability detection techniques, such as static and dynamic analysis, have been shown to be ineffective at detecting multiple vulnerabilities. To address this issue, this study employed a deep learning approach, specifically Convolutional Neural Networks (CNN), to solve the software vulnerability detection problem. A 5-split cross-validation approach was used to train and evaluate the CNN model, which takes tokenized source code as input. The findings indicated that Sherlock successfully detected multiple vulnerabilities at the function level, and its performance was particularly strong for CWE-199, CWE-120, and CWE-Other, with an overall high accuracy rate and significant true positive and true negative values. However, the performance was less reliable for some vulnerabilities due to the lack of a standardized dataset which will be a future research direction. The results suggest that compared to current techniques, the proposed deep learning approach has the potential to substantially enhance the accuracy of software vulnerability detection.

Prompt injection attacks can compromise the security and stability of critical systems, from infrastructure to large web applications. This work curates and augments a prompt injection dataset based on the HackAPrompt Playground Submissions corpus and trains several classifiers, including LSTM, feed forward neural networks, Random Forest, and Naive Bayes, to detect malicious prompts in LLM integrated web applications. The proposed approach improves prompt injection detection and mitigation, helping protect targeted applications and systems.

This article introduces a novel cryptographic paradigm based on nonderived polyadic algebraic structures. Traditional cryptosystems rely on binary operations within groups, rings, or fields, whose well-understood properties can be exploited in cryptanalysis. To overcome these vulnerabilities, we propose a shift to polyadic rings, which generalize classical rings by allowing operations of higher arity: an $m$-ary addition and an $n$-ary multiplication. The foundation of our approach is the construction of polyadic integers – congruence classes of ordinary integers endowed with such $m$-ary and $n$-ary operations. A key innovation is the parameter-to-arity mapping $\Phi(a,b)=(m,n)$, which links the parameters $(a,b)$ defining a congruence class to the specific arities required for algebraic closure. This mapping is mathematically intricate: it is non-injective, non-surjective, and multivalued. This complex, non-unique relationship forms the core of the proposed cryptosystem's security. We present two concrete encryption procedures that leverage this structure by encoding plaintext within the parameters of polyadic rings and transmitting information via polyadically quantized analog signals. In one method, plaintext is linked to the additive arity $m_{i}$ and secured using the summation of such signals; in the other, it is linked to a ring parameter $a_{i}$ and secured using their multiplication. In both cases, the "quantized" nature of polyadic operations generates systems of equations that are straightforward for a legitimate recipient with the correct key but exceptionally difficult for an attacker without it. The resulting framework promises a substantial increase in cryptographic security. This work establishes the theoretical foundation for this new class of encryption schemes and highlights their potential for constructing robust, next-generation cryptographic protocols.

The Federated Byzantine Agreement (FBA) achieves rapid consensus by relying on overlapping quorum slices. But this architecture leads to a high dependence on the availability of validators when about one fourth of validators go down, the classical FBA can lose liveness or fail to reach agreement. We thus come up with an Adaptive FBA architecture that can reconfigure quorum slices intelligently based on real time validator reputation to overcome this drawback. Our model includes trust scores computed from EigenTrust and a sliding window behavioral assessment to determine the reliability of validators. We have built the intelligent adaptive FBA model and conducted tests in a Stellar based setting. Results of real life experiments reveal that the system is stable enough to keep consensus when more than half of the validators (up to 62 percent) are disconnected, which is a great extension of the failure threshold of a classical FBA. A fallback mode allows the network to be functional with as few as three validators, thus showing a significant robustness enhancement. Besides, a comparative study with the existing consensus protocols shows that Adaptive FBA can be an excellent choice for the next generation of blockchain systems, especially for constructing a resilient blockchain infrastructure.

The widespread adoption of open-source ecosystems enables developers to integrate third-party packages, but also exposes them to malicious packages crafted to execute harmful behavior via public repositories such as PyPI. Existing datasets (e.g., pypi-malregistry, DataDog, OpenSSF, MalwareBench) label packages as malicious or benign at the package level, but do not specify which statements implement malicious behavior. This coarse granularity limits research and practice: models cannot be trained to localize malicious code, detectors cannot justify alerts with code-level evidence, and analysts cannot systematically study recurring malicious indicators or attack chains. To address this gap, we construct a statement-level dataset of 370 malicious Python packages (833 files, 90,527 lines) with 2,962 labeled occurrences of malicious indicators. From these annotations, we derive a fine-grained taxonomy of 47 malicious indicators across 7 types that capture how adversarial behavior is implemented in code, and we apply sequential pattern mining to uncover recurring indicator sequences that characterize common attack workflows. Our contribution enables explainable, behavior-centric detection and supports both semantic-aware model training and practical heuristics for strengthening software supply-chain defenses.

Safety and security are critical to the reliable operation of connected and automated vehicles (CAVs). While existing research has identified correlations between the two domains, a theoretical framework to analyze their interaction mechanisms and guide co-design remains lacking. To address this gap, this paper proposes a copula-based joint safety-security analysis method to quantify their coupling effects. First, we formulate time-varying cyberattacks using dynamic risk functions derived from survival analysis, while modeling random hardware failures with the Weibull distribution, as per the automotive industry standard ISO 26262. Second, to capture the dependence between functional safety failures and cyber threats, we introduce a joint failure model based on copula theory, employing both elliptical (e.g., Gaussian) and Archimedean (e.g., Frank) copula families to construct a system-level failure function. Furthermore, we provide formal theoretical analysis of the dependence structure in the safety-security coupling, yielding three key insights: (1) a monotonic relationship between joint failure probability and dependence parameters, (2) the mechanisms of defensive response mechanisms (such as patch deployment) in mitigating joint failures, and (3) quantifying the dynamic coupling strength between safety and security under dependence structures. Through comprehensive simulations, we evaluate the sensitivity of the joint failure behavior to three critical factors: copula dependence parameters, security patch deployment timing, and Weibull distribution parameters. Our dynamic failure model further illustrates how cyberattacks affect safety failures and, conversely, how functional faults affect security failures under dependencies structures. This study provides a quantifiable theoretical foundation for the co-design of safety and security in CAVs.

With the advent of machine learning and quantum computing, the 21st century has gone from a place of relative algorithmic security, to one of speculative unease and possibly, cyber catastrophe. Modern algorithms like Elliptic Curve Cryptography (ECC) are the bastion of current cryptographic security protocols that form the backbone of consumer protection ranging from Hypertext Transfer Protocol Secure (HTTPS) in the modern internet browser, to cryptographic financial instruments like Bitcoin. And there's been very little work put into testing the strength of these ciphers. Practically the only study that I could find was on side-channel recognition, a joint paper from the University of Milan, Italy and King's College, London\cite{battistello2025ecc}. These algorithms are already considered bulletproof by many consumers, but exploits already exist for them, and with computing power and distributed, federated compute on the rise, it's only a matter of time before these current bastions fade away into obscurity, and it's on all of us to stand up when we notice something is amiss, lest we see such passages claim victims in that process. In this paper, we seek to explore the use of modern language model architecture in cracking the association between a known public key, and its associated private key, by intuitively learning to reverse engineer the public keypair generation process, effectively solving the curve. Additonally, we attempt to ascertain modern machine learning's ability to memorize public-private secp256r1 keypairs, and to then test their ability to reverse engineer the public keypair generation process. It is my belief that proof-for would be equally valuable as proof-against in either of these categories. Finally, we'll conclude with some number crunching on where we see this particular field heading in the future.

The purpose of this study is to investigate how homophily, memory constraints, and adversarial disruptions collectively shape the resilience and adaptability of complex networks. To achieve this, we develop a new framework that integrates explicit memory decay mechanisms into homophily-based models and systematically evaluate their performance across diverse graph structures and adversarial settings. Our methods involve extensive experimentation on synthetic datasets, where we vary decay functions, reconnection probabilities, and similarity measures, primarily comparing cosine similarity with traditional metrics such as Jaccard similarity and baseline edge weights. The results show that cosine similarity achieves up to a 30% improvement in stability metrics in sparse, convex, and modular networks. Moreover, the refined value-of-recall metric demonstrates that strategic forgetting can bolster resilience by balancing network robustness and adaptability. The findings underscore the critical importance of aligning memory and similarity parameters with the structural and adversarial dynamics of the network. By quantifying the tangible benefits of incorporating memory constraints into homophily-based analyses, this study offers actionable insights for optimizing real-world applications, including social systems, collaborative platforms, and cybersecurity contexts.

The rapid proliferation of Artificial Intelligence Generated Content has precipitated a crisis of trust and urgent regulatory demands. However, existing identification tools suffer from fragmentation and a lack of support for visible compliance marking. To address these gaps, we introduce the UniMark, an open-source, unified framework for multimodal content governance. Our system features a modular unified engine that abstracts complexities across text, image, audio, and video modalities. Crucially, we propose a novel dual-operation strategy, natively supporting both Hidden Watermarking for copyright protection and Visible Marking for regulatory compliance. Furthermore, we establish a standardized evaluation framework with three specialized benchmarks (Image/Video/Audio-Bench) to ensure rigorous performance assessment. This toolkit bridges the gap between advanced algorithms and engineering implementation, fostering a more transparent and secure digital ecosystem.

Software supply chain attacks on the npm ecosystem have grown increasingly sophisticated, exploiting obfuscation and complex logic to evade detection. Large Language Models (LLMs) offer strong semantic understanding of code but face practical constraints: limited context windows and high inference costs make full-package analysis infeasible, while naive token-based splitting fragments semantic context and degrades accuracy. This paper introduces an LLM-based framework for malicious npm package detection built on code-slicing techniques. We propose an adaptation of taint-based slicing for the npm ecosystem, guided by a curated inventory of JavaScript-specific sensitive APIs, to isolate security-relevant data flows from benign boilerplate. The approach reduces the mean input token count by 99.75% and the median by 93.7% while preserving critical malicious behaviors. Packages relying on dynamic code generation or obfuscation yield empty slices under static analysis and require deobfuscation preprocessing, a limitation we explicitly discuss. The framework is evaluated on a dataset of more than 7000 malicious and benign npm packages using DeepSeek-Coder6.7B. On the 2537 packages amenable to static taint analysis, taint-based slicing achieves 87.04% detection accuracy, outperforming both a naive token-splitting baseline at 75.41% and a CFG-only static slicing approach at 75.65%. These results demonstrate that semantically targeted input representations improve LLM-based detection performance beyond what is achievable through simple input-size reduction, providing an effective and computationally practical defense against evolving open-source supply-chain threats.

The increasing use of machine learning (ML) for Just-In-Time (JIT) defect prediction raises concerns about privacy leakage from software analytics data. Existing anonymization methods, such as tabular transformations and graph perturbations, often overlook contextual dependencies among software metrics, leading to suboptimal privacy-utility tradeoffs. Leveraging the contextual reasoning of Large Language Models (LLMs), we propose a cluster-guided anonymization technique that preserves contextual and statistical relationships within JIT datasets. Our method groups commits into feature-based clusters and employs an LLM to generate context-aware parameter configurations for each commit cluster, defining alpha-beta ratios and churn mixture distributions used for anonymization. Our evaluation on six projects (Cassandra, Flink, Groovy, Ignite, OpenStack, and Qt) shows that our LLM-based approach achieves privacy level 2 (IPR >= 80 percent), improving privacy by 18 to 25 percent over four state-of-the-art graph-based anonymization baselines while maintaining comparable F1 scores. Our results demonstrate that LLMs can act as adaptive anonymization engines when provided with cluster-specific statistical information about similar data points, enabling context-sensitive and privacy-preserving software analytics without compromising predictive accuracy.

EIP-7702 introduces a delegation-based authorization mechanism that allows an externally owned account (EOA) to authenticate a single authorization tuple, after which all subsequent calls are routed to arbitrary delegate code. We show that this design enables a qualitatively new class of phishing attacks: instead of deceiving users into signing individual transactions, an attacker can induce a victim to sign a single authorization tuple that grants unconditional and persistent execution control over the account. Through controlled experiments, we identify three reliable trigger pathways: user-driven, attacker-driven, and protocol-triggered. Each can lead to full account takeover and complete asset drainage. We further propose two extended attack surfaces. First, ERC-4337's EntryPoint pipeline enables remote and repeated activation of the delegated code without further victim involvement. Second, the chain-agnostic authorization mode permits replay-like compromises across independent networks. We also present the first empirical measurement of EIP-7702 usage across major EVM chains. Analyzing over 150k authorization and execution events involving 26k addresses and hundreds of delegator contracts, we assess the protocol's real-world footprint. Our findings show that EIP-7702 authorizations are highly centralized, dominated by a small number of contract families linked to criminal activity and repeatedly reused across incidents. Corresponding loss data reveals substantial theft of ETH, ERC-20 tokens, and NFTs. These results provide practical evidence that the attack surface we identify is not merely theoretical, but is already being exploited at scale. We conclude by proposing protocol-level defenses to mitigate the delegation-based phishing vector introduced by EIP-7702.

Accurate time-series forecasting is increasingly critical for planning and operations in low-carbon power systems. Emerging time-series large language models (TS-LLMs) now deliver this capability at scale, requiring no task-specific retraining, and are quickly becoming essential components within the Internet-of-Energy (IoE) ecosystem. However, their real-world deployment is complicated by a critical vulnerability: adversarial examples (AEs). Detecting these AEs is challenging because (i) adversarial perturbations are optimized across the entire input sequence and exploit global temporal dependencies, which renders local detection methods ineffective, and (ii) unlike traditional forecasting models with fixed input dimensions, TS-LLMs accept sequences of variable length, increasing variability that complicates detection. To address these challenges, we propose a plug-in detection framework that capitalizes on the TS-LLM's own variable-length input capability. Our method uses sampling-induced divergence as a detection signal. Given an input sequence, we generate multiple shortened variants and detect AEs by measuring the consistency of their forecasts: Benign sequences tend to produce stable predictions under sampling, whereas adversarial sequences show low forecast similarity, because perturbations optimized for a full-length sequence do not transfer reliably to shorter, differently-structured subsamples. We evaluate our approach on three representative TS-LLMs (TimeGPT, TimesFM, and TimeLLM) across three energy datasets: ETTh2 (Electricity Transformer Temperature), NI (Hourly Energy Consumption), and Consumption (Hourly Electricity Consumption and Production). Empirical results confirm strong and robust detection performance across both black-box and white-box attack scenarios, highlighting its practicality as a reliable safeguard for TS-LLM forecasting in real-world energy systems.

Industry 5.0's increasing integration of IT and OT systems is transforming industrial operations but also expanding the cyber-physical attack surface. Industrial Control Systems (ICS) face escalating security challenges as traditional siloed defences fail to provide coherent, cross-domain threat insights. We present BRIDG-ICS (BRIDge for Industrial Control Systems), an AI-driven Knowledge Graph (KG) framework for context-aware threat analysis and quantitative assessment of cyber resilience in smart manufacturing environments. BRIDG-ICS fuses heterogeneous industrial and cybersecurity data into an integrated Industrial Security Knowledge Graph linking assets, vulnerabilities, and adversarial behaviours with probabilistic risk metrics (e.g. exploit likelihood, attack cost). This unified graph representation enables multi-stage attack path simulation using graph-analytic techniques. To enrich the graph's semantic depth, the framework leverages Large Language Models (LLMs): domain-specific LLMs extract cybersecurity entities, predict relationships, and translate natural-language threat descriptions into structured graph triples, thereby populating the knowledge graph with missing associations and latent risk indicators. This unified AI-enriched KG supports multi-hop, causality-aware threat reasoning, improving visibility into complex attack chains and guiding data-driven mitigation. In simulated industrial scenarios, BRIDG-ICS scales well, reduces potential attack exposure, and can enhance cyber-physical system resilience in Industry 5.0 settings.