Papers

Multimodal learning has shown significant superiority on various tasks by integrating multiple modalities. However, the interdependencies among modalities increase the susceptibility of multimodal models to adversarial attacks. Existing methods mainly focus on attacks on specific modalities or indiscriminately attack all modalities. In this paper, we find that these approaches ignore the differences between modalities in their contribution to final robustness, resulting in suboptimal robustness performance. To bridge this gap, we introduce Vulnerability-Aware Robust Multimodal Adversarial Training (VARMAT), a probe-in-training adversarial training method that improves multimodal robustness by identifying the vulnerability of each modality. To be specific, VARMAT first explicitly quantifies the vulnerability of each modality, grounded in a first-order approximation of the attack objective (Probe). Then, we propose a targeted regularization term that penalizes modalities with high vulnerability, guiding robust learning while maintaining task accuracy (Training). We demonstrate the enhanced robustness of our method across multiple multimodal datasets involving diverse modalities. Finally, we achieve {12.73%, 22.21%, 11.19%} robustness improvement on three multimodal datasets, revealing a significant blind spot in multimodal adversarial training.

Securing AI agents powered by Large Language Models (LLMs) represents one of the most critical challenges in AI security today. Unlike traditional software, AI agents leverage LLMs as their "brain" to autonomously perform actions via connected tools. This capability introduces significant risks that go far beyond those of harmful text presented in a chatbot that was the main application of LLMs. A compromised AI agent can deliberately abuse powerful tools to perform malicious actions, in many cases irreversible, and limited solely by the guardrails on the tools themselves and the LLM ability to enforce them. This paper presents ASTRA, a first-of-its-kind framework designed to evaluate the effectiveness of LLMs in supporting the creation of secure agents that enforce custom guardrails defined at the system-prompt level (e.g., "Do not send an email out of the company domain," or "Never extend the robotic arm in more than 2 meters"). Our holistic framework simulates 10 diverse autonomous agents varying between a coding assistant and a delivery drone equipped with 37 unique tools. We test these agents against a suite of novel attacks developed specifically for agentic threats, inspired by the OWASP Top 10 but adapted to challenge the ability of the LLM for policy enforcement during multi-turn planning and execution of strict tool activation. By evaluating 13 open-source, tool-calling LLMs, we uncovered surprising and significant differences in their ability to remain secure and keep operating within their boundaries. The purpose of this work is to provide the community with a robust and unified methodology to build and validate better LLMs, ultimately pushing for more secure and reliable agentic AI systems.

This paper presents an empirical investigation into the capabilities of Large Language Models (LLMs) to perform automated Attribute-based Access Control (ABAC) policy mining. While ABAC provides fine-grained, context-aware access management, the increasing number and complexity of access policies can make their formulation and evaluation rather challenging. To address the task of synthesizing concise yet accurate policies, we evaluate the performance of some of the state-of-the-art LLMs, specifically Google Gemini (Flash and Pro) and OpenAI ChatGPT, as potential policy mining engines. An experimental framework was developed in Python to generate randomized access data parameterized by varying numbers of subjects, objects, and initial policy sets. The baseline policy sets, which govern permission decisions between subjects and objects, serve as the ground truth for comparison. Each LLM-generated policy was evaluated against the baseline policy using standard performance metrics. The results indicate that LLMs can effectively infer compact and valid ABAC policies for small-scale scenarios. However, as the system size increases, characterized by higher numbers of subjects and objects, LLM outputs exhibit declining accuracy and precision, coupled with significant increase in the size of policy generated, which is beyond the optimal size. These findings highlight both the promise and limitations of current LLM architectures for scalable policy mining in access control domains. Future work will explore hybrid approaches that combine prompt optimization with classical rule mining algorithms to improve scalability and interpretability in complex ABAC environments.

The exponential growth of the Internet of Things (IoT) ecosystem has amplified concerns regarding device reliability, interoperability, and security assurance. Despite the proliferation of IoT security guidelines, a unified and quantitative approach to measuring trust remains absent. This paper introduces SCI-IoT (Secure Certification Index for IoT), a standardized and quantitative framework for trust scoring, evaluation, and certification of IoT devices. The framework employs a six-tier grading model (Grades A-F), enabling device profiling across consumer, industrial, and critical infrastructure domains. Within this model, 30 distinct Trust Tests assess devices across dimensions such as authentication, encryption, data integrity, resilience, and firmware security. Each test is assigned a criticality-based weight (1.0-2.0) and a performance rating (1-4), converted to a normalized percentage and aggregated through a weighted computation to yield the Secure Certification Index (SCI). The SCI determines the device's Trust Verdict, categorized into five SCI levels, and serves as the foundation for optional grade-based certification. The framework also incorporates critical gate conditions, enforcing absolute compliance in high risk parameters to prevent certification of devices with fundamental vulnerabilities. By unifying quantitative trust scoring with structured certification criteria, SCI-IoT provides a transparent, scalable, and reproducible method to benchmark IoT device trustworthiness. The proposed system aims to streamline manufacturer compliance, improve consumer confidence, and facilitate global interoperability in IoT security certification.

Data streams collected from multiple sources are rarely independent. Values evolve over time and influence one another across sequences. These correlations improve prediction in healthcare, finance, and smart-city control yet violate the record-independence assumption built into most Differential Privacy (DP) mechanisms. To restore rigorous privacy guarantees without sacrificing utility, we introduce Correlated-Sequence Differential Privacy (CSDP), a framework specifically designed for preserving privacy in correlated sequential data. CSDP addresses two linked challenges: quantifying the extra information an attacker gains from joint temporal and cross-sequence links, and adding just enough noise to hide that information while keeping the data useful. We model multivariate streams as a Coupling Markov Chain, yielding the derived loose leakage bound expressed with a few spectral terms and revealing a counterintuitive result: stronger coupling can actually decrease worst-case leakage by dispersing perturbations across sequences. Guided by these bounds, we build the Freshness-Regulated Adaptive Noise (FRAN) mechanism–combining data aging, correlation-aware sensitivity scaling, and Laplace noise–that runs in linear time. Tests on two-sequence datasets show that CSDP improves the privacy-utility trade-off by approximately 50% over existing correlated-DP methods and by two orders of magnitude compared to the standard DP approach.

Multi-domain graph pre-training has emerged as a pivotal technique in developing graph foundation models. While it greatly improves the generalization of graph neural networks, its privacy risks under membership inference attacks (MIAs), which aim to identify whether a specific instance was used in training (member), remain largely unexplored. However, effectively conducting MIAs against multi-domain graph pre-trained models is a significant challenge due to: (i) Enhanced Generalization Capability: Multi-domain pre-training reduces the overfitting characteristics commonly exploited by MIAs. (ii) Unrepresentative Shadow Datasets: Diverse training graphs hinder the obtaining of reliable shadow graphs. (iii) Weakened Membership Signals: Embedding-based outputs offer less informative cues than logits for MIAs. To tackle these challenges, we propose MGP-MIA, a novel framework for Membership Inference Attacks against Multi-domain Graph Pre-trained models. Specifically, we first propose a membership signal amplification mechanism that amplifies the overfitting characteristics of target models via machine unlearning. We then design an incremental shadow model construction mechanism that builds a reliable shadow model with limited shadow graphs via incremental learning. Finally, we introduce a similarity-based inference mechanism that identifies members based on their similarity to positive and negative samples. Extensive experiments demonstrate the effectiveness of our proposed MGP-MIA and reveal the privacy risks of multi-domain graph pre-training.

Graph Foundation Models (GFMs) are pre-trained on diverse source domains and adapted to unseen targets, enabling broad generalization for graph machine learning. Despite that GFMs have attracted considerable attention recently, their vulnerability to backdoor attacks remains largely underexplored. A compromised GFM can introduce backdoor behaviors into downstream applications, posing serious security risks. However, launching backdoor attacks against GFMs is non-trivial due to three key challenges. (1) Effectiveness: Attackers lack knowledge of the downstream task during pre-training, complicating the assurance that triggers reliably induce misclassifications into desired classes. (2) Stealthiness: The variability in node features across domains complicates trigger insertion that remains stealthy. (3) Persistence: Downstream fine-tuning may erase backdoor behaviors by updating model parameters. To address these challenges, we propose GFM-BA, a novel Backdoor Attack model against Graph Foundation Models. Specifically, we first design a label-free trigger association module that links the trigger to a set of prototype embeddings, eliminating the need for knowledge about downstream tasks to perform backdoor injection. Then, we introduce a node-adaptive trigger generator, dynamically producing node-specific triggers, reducing the risk of trigger detection while reliably activating the backdoor. Lastly, we develop a persistent backdoor anchoring module that firmly anchors the backdoor to fine-tuning-insensitive parameters, enhancing the persistence of the backdoor under downstream adaptation. Extensive experiments demonstrate the effectiveness, stealthiness, and persistence of GFM-BA.

Electric Vehicle (EV) charging infrastructure faces escalating cybersecurity threats that can severely compromise operational efficiency and grid stability. Existing forecasting techniques are limited by the lack of combined robust anomaly mitigation solutions and data privacy preservation. Therefore, this paper addresses these challenges by proposing a novel anomaly-resilient federated learning framework that simultaneously preserves data privacy, detects cyber-attacks, and maintains trustworthy demand prediction accuracy under adversarial conditions. The proposed framework integrates three key innovations: LSTM autoencoder-based distributed anomaly detection deployed at each federated client, interpolation-based anomalous data mitigation to preserve temporal continuity, and federated Long Short-Term Memory (LSTM) networks that enable collaborative learning without centralized data aggregation. The framework is validated on real-world EV charging infrastructure datasets combined with real-world DDoS attack datasets, providing robust validation of the proposed approach under realistic threat scenarios. Experimental results demonstrate that the federated approach achieves superior performance compared to centralized models, with 15.2% improvement in R2 accuracy while maintaining data locality. The integrated cyber-attack detection and mitigation system produces trustworthy datasets that enhance prediction reliability, recovering 47.9% of attack-induced performance degradation while maintaining exceptional precision (91.3%) and minimal false positive rates (1.21%). The proposed architecture enables enhanced EV infrastructure planning, privacy-preserving collaborative forecasting, cybersecurity resilience, and rapid recovery from malicious threats across distributed charging networks.

Maintaining economic efficiency and operational reliability in microgrid energy management systems under cyberattack conditions remains challenging. Most approaches assume non-anomalous measurements, make predictions with unquantified uncertainties, and do not mitigate malicious attacks on renewable forecasts for energy management optimization. This paper presents a comprehensive cyber-resilient framework integrating federated Long Short-Term Memory-based photovoltaic forecasting with a novel two-stage cascade false data injection attack detection and energy management system optimization. The approach combines autoencoder reconstruction error with prediction uncertainty quantification to enable attack-resilient energy storage scheduling while preserving data privacy. Extreme false data attack conditions were studied that caused 58% forecast degradation and 16.9% operational cost increases. The proposed integrated framework reduced false positive detections by 70%, recovered 93.7% of forecasting performance losses, and achieved 5% operational cost savings, mitigating 34.7% of attack-induced economic losses. Results demonstrate that precision-focused cascade detection with multi-signal fusion outperforms single-signal approaches, validating security-performance synergy for decentralized microgrids.

As AI agents attempt to autonomously act on users' behalf, they raise transparency and control issues. We argue that permission-based access control is indispensable in providing meaningful control to the users, but conventional permission models are inadequate for the automated agentic execution paradigm. We therefore propose automated permission management for AI agents. Our key idea is to conduct a user study to identify the factors influencing users' permission decisions and to encode these factors into an ML-based permission management assistant capable of predicting users' future decisions. We find that participants' permission decisions are influenced by communication context but importantly individual preferences tend to remain consistent within contexts, and align with those of other participants. Leveraging these insights, we develop a permission prediction model achieving 85.1% accuracy overall and 94.4% for high-confidence predictions. We find that even without using permission history, our model achieves an accuracy of 66.9%, and a slight increase of training samples (i.e., 1-4) can substantially increase the accuracy by 10.8%.

LLM applications (i.e., LLM apps) leverage the powerful capabilities of LLMs to provide users with customized services, revolutionizing traditional application development. While the increasing prevalence of LLM-powered applications provides users with unprecedented convenience, it also brings forth new security challenges. For such an emerging ecosystem, the security community lacks sufficient understanding of the LLM application ecosystem, especially regarding the capability boundaries of the applications themselves. In this paper, we systematically analyzed the new development paradigm and defined the concept of the LLM app capability space. We also uncovered potential new risks beyond jailbreak that arise from ambiguous capability boundaries in real-world scenarios, namely, capability downgrade and upgrade. To evaluate the impact of these risks, we designed and implemented an LLM app capability evaluation framework, LLMApp-Eval. First, we collected application metadata across 4 platforms and conducted a cross-platform ecosystem analysis. Then, we evaluated the risks for 199 popular applications among 4 platforms and 6 open-source LLMs. We identified that 178 (89.45%) potentially affected applications, which can perform tasks from more than 15 scenarios or be malicious. We even found 17 applications in our study that executed malicious tasks directly, without applying any adversarial rewriting. Furthermore, our experiments also reveal a positive correlation between the quality of prompt design and application robustness. We found that well-designed prompts enhance security, while poorly designed ones can facilitate abuse. We hope our work inspires the community to focus on the real-world risks of LLM applications and foster the development of a more robust LLM application ecosystem.

Privacy concerns around 5G, the latest generation of mobile networks, are growing, with fears that its deployment may increase exposure to privacy risks. This perception is largely driven by the use of denser deployments of small antenna systems, which enable highly accurate data collection at higher speeds and closer proximity to mobile users. At the same time, 5G's unique radio communication features can make the reproduction of known network attacks more challenging. In particular, passive network attacks, which do not involve direct interaction with the target network and are therefore nearly impossible to detect, remain a pressing concern. Such attacks can reveal sensitive information about users, their devices, and active applications, which may then be exploited through known vulnerabilities or spear-phishing schemes. This survey examines the feasibility of passive network attacks in 5G and beyond (B5G/6G) networks, with emphasis on two major categories: information extraction (system identification, website and application fingerprinting) and geolocation (user identification and position tracking). These attacks are well documented and reproducible in existing wireless and mobile systems, including short-range networks (IEEE 802.11) and, to a lesser extent, LTE. Current evidence suggests that while such attacks remain theoretically possible in 5G, their practical execution is significantly constrained by directional beamforming, high-frequency propagation characteristics, and encryption mechanisms. For B5G and early 6G networks, the lack of public tools and high hardware cost currently renders these attacks infeasible in practice, which highlights a critical gap in our understanding of future network threat models.

SOCs and CSIRTs face increasing pressure to automate incident categorization, yet the use of cloud-based LLMs introduces costs, latency, and confidentiality risks. We investigate whether locally executed SLMs can meet this challenge. We evaluated 21 models ranging from 1B to 20B parameters, varying the temperature hyperparameter and measuring execution time and precision across two distinct architectures. The results indicate that temperature has little influence on performance, whereas the number of parameters and GPU capacity are decisive factors.

In this work, we present homomorphic encryption-based vaults (Haults), a permissioned privacy-preserving smart wallet protocol for VM-enabled blockchains that keeps users' balances confidential, as well as the amounts transacted to other parties. To comply with regulations, we include optional compliance features that allow specific entities (the auditors) to retrieve transaction amounts or execute force transfers when necessary. Our solution uses ElGamal over elliptic curves to encrypt balances, combined with zero-knowledge proofs to verify the correctness of transaction amounts and the integrity of the sender's updated balance, among other security checks. We provide a detailed explanation of the protocol, including a security discussion and benchmarks from our proof-of-concept implementation, which yield great results. Beyond in-contract issued tokens, we also provide a thorough explanation on how our solution can be compatible with external ones (e.g., Ether or any ERC20).

Security bugs in the Linux kernel emerge endlessly and have attracted much attention. However, fixing security bugs in the Linux kernel could be incomplete due to human mistakes. Specifically, an incomplete fix fails to repair all the original security defects in the software, fails to properly repair the original security defects, or introduces new ones. In this paper, we study the fixes of incomplete security bugs in the Linux kernel for the first time, and reveal their characteristics, root causes, as well as detection. We first construct a dataset of incomplete security bug fixes in the Linux kernel and answer the following three questions. What are the characteristics of incomplete security bug fixes in the Linux kernel? What are the root causes behind them? How should they be detected to reduce security risks? We then have the three main insights in the following. (*Due to the notification of arXiv "The Abstract field cannot be longer than 1,920 characters", the appeared Abstract is shortened. For the full Abstract, please download the Article.)

Memory-safety escapes continue to form the launching pad for a wide range of security attacks, especially for the substantial base of deployed software that is coded in pointer-based languages such as C/C++. Although compiler and Instruction Set Architecture (ISA) extensions have been introduced to address elements of this issue, the overhead and/or comprehensive applicability have limited broad production deployment. The Memory Tagging Extension (MTE) to the ARM AArch64 Instruction Set Architecture is a valuable tool to address memory-safety escapes; when used in synchronous tag-checking mode, MTE provides deterministic detection and prevention of sequential buffer overflow attacks, and probabilistic detection and prevention of exploits resulting from temporal use-after-free pointer programming bugs. The AmpereOne processor, launched in 2024, is the first datacenter processor to support MTE. Its optimized MTE implementation uniquely incurs no memory capacity overhead for tag storage and provides synchronous tag-checking with single-digit performance impact across a broad range of datacenter class workloads. Furthermore, this paper analyzes the complete hardware-software stack, identifying application memory management as the primary remaining source of overhead and highlighting clear opportunities for software optimization. The combination of an efficient hardware foundation and a clear path for software improvement makes the MTE implementation of the AmpereOne processor highly attractive for deployment in production cloud environments.

Intrusion Detection Systems (IDS) are critical to defending enterprise and industrial control environments, yet evaluating their effectiveness under realistic conditions remains an open challenge. Existing benchmarks rely on synthetic datasets (e.g., NSL-KDD, CICIDS2017) or scripted replay frameworks, which fail to capture adaptive adversary behavior. Even MITRE ATT&CK Evaluations, while influential, are host-centric and assume malware-driven compromise, thereby under-representing stealthy, multi-stage intrusions across IT and OT domains. We present StealthCup, a novel evaluation methodology that operationalizes IDS benchmarking as an evasion-focused Capture-the-Flag competition. Professional penetration testers engaged in multi-stage attack chains on a realistic IT/OT testbed, with scoring penalizing IDS detections. The event generated structured attacker writeups, validated detections, and PCAPs, host logs, and alerts. Our results reveal that out of 32 exercised attack techniques, 11 were not detected by any IDS configuration. Open-source systems (Wazuh, Suricata) produced high false-positive rates >90%, while commercial tools generated fewer false positives but also missed more attacks. Comparison with the Volt Typhoon APT advisory confirmed strong realism: all 28 applicable techniques were exercised, 19 appeared in writeups, and 9 in forensic traces. These findings demonstrate that StealthCup elicits attacker behavior closely aligned with state-sponsored TTPs, while exposing blind spots across both open-source and commercial IDS. The resulting datasets and methodology provide a reproducible foundation for future stealth-focused IDS evaluation.

Flexible energy resources are increasingly becoming common in smart grids. These resources are typically managed and controlled by aggregators that coordinate many resources to provide flexibility services. However, these aggregators and flexible energy resources are vulnerable, which could allow attackers to remotely control flexible energy resources to launch large-scale attacks on the grid. This paper investigates and evaluates the potential attack strategies that can be used to manipulate flexible energy resources to challenge the effectiveness of traditional grid stability measures and disrupt the first-swing stability of the power grid. Our work shows that although a large amount of power is required, the current flexibility capacities could potentially be sufficient to disrupt the grid on a national level.

Recent work has shown that out-of-order and speculative execution mechanisms used to increase performance in the majority of processors expose the processors to critical attacks. These attacks, called Meltdown and Spectre, exploit the side effects of performance-enhancing features in modern microprocessors to expose secret data through side channels in the microarchitecture. The well known implementations of these attacks exploit cache-based side channels since they are the least noisy channels to exfiltrate data. While some software patches attempted to mitigate these attacks, they are ad-hoc and only try to fix the side effects of the vulnerabilites. They may also impose a performance overhead of up to 30%. In this paper, we present a microarchitecture-based solution for Meltdown and Spectre that addresses the vulnerabilities exploited by the attacks. Our solution prevents flushed instructions from exposing data to the cache. Our approach can also be extended to other memory structures in the microarchitecture thereby preventing variants of the attacks which exploit these memory structures. We further identify two new variant attacks based on exploiting the side effects of speculative and out-of-order execution and show how our solution can be used to prevent these attacks. Evaluation results show that our microarchitectural solution not only restores secure out-of-order and speculative execution, but also has relatively low overhead and does not significantly impact performance for most applications.

QDNA-ID is a trust-chain framework that links physical quantum behavior to digitally verified records. The system first executes standard quantum circuits with random shot patterns across different devices to generate entropy profiles and measurement data that reveal device-specific behavior. A Bell or CHSH test is then used to confirm that correlations originate from genuine non classical processes rather than classical simulation. The verified outcomes are converted into statistical fingerprints using entropy, divergence, and bias features to characterize each device. These features and metadata for device, session, and random seed parameters are digitally signed and time stamped to ensure integrity and traceability. Authenticated artifacts are stored in a hierarchical index for reproducible retrieval and long term auditing. A visualization and analytics interface monitors drift, policy enforcement, and device behavior logs. A machine learning engine tracks entropy drift, detects anomalies, and classifies devices based on evolving patterns. An external verification API supports independent recomputation of hashes, signatures, and CHSH evidence. QDNA-ID operates as a continuous feedback loop that maintains a persistent chain of trust for quantum computing environments.

Language agents are moving from single-user assistants to shared collaborators in workspaces, forums, and group chats. A shared agent observes interleaved messages from multiple users, maintains persistent context, and executes tool actions, creating a boundary problem absent from standard single-user agents: the agent must decide not only whether an instruction is safe, but which user or task it is allowed to govern. We identify cross-user poisoning (CUP), where an adversary injects a message into shared context that is later applied while the agent serves a benign user, causing unauthorized actions or responses outside the instruction's intended scope. We validate CUP on two deployed multi-user agents, Continua and ElizaOS, and introduce MURMUR, a framework for evaluating shared-context agents under concurrent multi-user interactions. Across Slack, Workspace, and Airline domains, CUP achieves high attack success, persists across later interactions, and remains substantially more effective than matched prompt-injection attacks. We evaluate boundary-scoping defenses and find that context summarization alone is insufficient, while task clustering and provenance prompting substantially reduce non-adaptive propagation. These results show that robust multi-user agents require explicit user/task scoping rather than generic input filtering or context compression.

The recent boom and rapid integration of Large Language Models (LLMs) into a wide range of applications warrants a deeper understanding of their security and safety vulnerabilities. This paper presents a comparative analysis of the susceptibility to jailbreak attacks for two leading publicly available LLMs, Google's Gemini 2.5 Flash and OpenAI's GPT-4 (specifically the GPT-4o mini model accessible in the free tier). The research utilized two main bypass strategies: 'self-bypass', where models were prompted to circumvent their own safety protocols, and 'cross-bypass', where one model generated adversarial prompts to exploit vulnerabilities in the other. Four attack methods were employed - direct injection, role-playing, context manipulation, and obfuscation - to generate five distinct categories of unsafe content: hate speech, illegal activities, malicious code, dangerous content, and misinformation. The success of the attack was determined by the generation of disallowed content, with successful jailbreaks assigned a severity score. The findings indicate a disparity in jailbreak susceptibility between 2.5 Flash and GPT-4, suggesting variations in their safety implementations or architectural design. Cross-bypass attacks were particularly effective, indicating that an ample amount of vulnerabilities exist in the underlying transformer architecture. This research contributes a scalable framework for automated AI red-teaming and provides data-driven insights into the current state of LLM safety, underscoring the complex challenge of balancing model capabilities with robust safety mechanisms.

We present a patient-centric architecture for electronic health record (EHR) sharing that separates content storage from authorization and audit. Encrypted FHIR resources are stored off-chain; a public blockchain records only cryptographic commitments and patient-signed, time-bounded permissions using EIP-712. Keys are distributed via public-key wrapping, enabling storage providers to remain honest-but-curious without risking confidentiality. We formalize security goals (confidentiality, integrity, cryptographically attributable authorization, and auditability of authorization events) and provide a Solidity reference implementation deployed as single-patient contracts. On-chain costs for permission grants average 78,000 gas (L1), and end-to-end access latency for 1 MB records is 0.7–1.4s (mean values for S3 and IPFS respectively), dominated by storage retrieval. Layer-2 deployment reduces gas usage by 10–13x, though data availability charges dominate actual costs. We discuss metadata privacy, key registry requirements, and regulatory considerations (HIPAA/GDPR), demonstrating a practical route to restoring patient control while preserving security properties required for sensitive clinical data.

With the rapid growth of IoT, secure and efficient mesh networking has become essential. Thread has emerged as a key protocol, widely used in smart-home and commercial systems, and serving as a core transport layer in the Matter standard. This paper presents ThreadFuzzer, the first dedicated fuzzing framework for systematically testing Thread protocol implementations. By manipulating packets at the MLE layer, ThreadFuzzer enables fuzzing of both virtual OpenThread nodes and physical Thread devices. The framework incorporates multiple fuzzing strategies, including Random and Coverage-based fuzzers from CovFuzz, as well as a newly introduced TLV Inserter, designed specifically for TLV-structured MLE messages. These strategies are evaluated on the OpenThread stack using code-coverage and vulnerability-discovery metrics. The evaluation uncovered five previously unknown vulnerabilities in the OpenThread stack, several of which were successfully reproduced on commercial devices that rely on OpenThread. Moreover, ThreadFuzzer was benchmarked against an oracle AFL++ setup using the manually extended OSS-Fuzz harness from OpenThread, demonstrating strong effectiveness. These results demonstrate the practical utility of ThreadFuzzer while highlighting challenges and future directions in the wireless protocol fuzzing research space.

Private BitTorrent trackers enforce upload-to-download ratios to prevent free-riding, but suffer from three critical weaknesses: reputation cannot move between trackers, centralized servers create single points of failure, and upload statistics are self-reported and unverifiable. When a tracker shuts down, users lose their contribution history and cannot prove their standing to new communities. We address these problems by storing reputation in smart contracts and replacing self-reports with cryptographic attestations. Peers sign receipts for received pieces; the tracker aggregates them via BLS signatures and updates reputation. If a tracker is unavailable, peers fall back to an authenticated distributed hash table (DHT): stored reputation acts as a public key infrastructure (PKI), preserving access control without the tracker. Reputation is portable across tracker failures through single-hop migration in factory-deployed contracts. We also address the privacy implications of publishing public keys and reputations tied to private trackers on a public ledger: we propose ephemeral session keys to prevent linking peer identities, zero-knowledge membership proofs for anonymous DHT participation, and confidential reputation using homomorphic commitments. We formalize the security requirements, prove four security properties under standard cryptographic assumptions, and evaluate a prototype. Measurements show that transfer receipts add less than 5% end-to-end overhead with typical piece sizes. To minimize signing overhead, we adopt a hybrid signature scheme: ECDSA signs individual piece receipts at transfer time for low per-operation latency, while BLS serves as the overarching scheme, enabling compact aggregation of many receipts into a single proof at report time. This design reduces client-side signing cost by an order of magnitude compared to using BLS throughout.

We show that attention sinks and compression valleys create a vulnerable region in decoder-only Transformers, where small activation perturbations can be amplified through the autoregressive trajectory. Based on this, we propose Sensitivity-Scaled Steering (SSS), a progressive activation-space attack that anchors perturbations at the beginning-of-sequence token and adaptively reinforces them at sensitive layers and tokens. Instead of forcing an abrupt behavioral change, SSS induces a staged drift, making outputs gradually shift toward the target behavior while remaining fluent and benign-looking in early generations. Across multiple open-weight models and four behavioral axes, SSS achieves high attack success, preserves coherence, and causes negligible degradation to general capabilities. These results show that attention sinks and compression valleys are not merely mechanistic features; rather, they expose exploitable amplification mechanisms that can be treated as hidden-state weaknesses for activation-space attacks in white-box and supply-chain LLM deployments.

We investigate the problem of detecting dependencies between the components of a high-dimensional vector. Our approach advances the existing literature in two important respects. First, we consider the problem under privacy constraints. Second, instead of testing whether the coordinates are pairwise independent, we are interested in determining whether certain pairwise associations between the components (such as all pairwise Kendall's $\tau$ coefficients) do not exceed a given threshold in absolute value. Considering hypotheses of this form is motivated by the observation that in the high-dimensional regime, it is rare and perhaps impossible to have a null hypothesis that can be modeled exactly by assuming that all pairwise associations are precisely equal to zero. The formulation of the null hypothesis as a composite hypothesis makes the problem of constructing tests already non-standard in the non-private setting. Additionally, under privacy constraints, state of the art procedures rely on permutation approaches that are rendered invalid under a composite null. We propose a novel bootstrap based methodology that is especially powerful in sparse settings, develop theoretical guarantees under mild assumptions and show that the proposed method enjoys good finite sample properties even in the high privacy regime. Additionally, we present applications in medical data that showcase the applicability of our methodology.

Regulated AI workflows (such as clinical trials, medical decision support, and financial compliance) must satisfy strict auditability and integrity requirements. Existing audit-trail mechanisms rely on variable-length records, bulky cryptographic transcripts, or ad-hoc schemas, suffering from metadata leakage, irregular performance, and weak alignment with formal security notions.This paper introduces constant-size cryptographic evidence structures, a general abstraction for verifiable audit evidence in regulated AI workflows. Each evidence item is a fixed-size tuple of cryptographic fields designed to (i) bind strongly to workflow events and configurations, (ii) support constant-size storage and uniform verification cost per event, and (iii) compose cleanly with hash-chain and Merkle-based audit constructions. We formalize a model of regulated AI workflows, define syntax and algorithms for evidence structures, and prove security properties (evidence binding, tamper detection, and non-equivocation) via game-based definitions under standard assumptions (collision-resistant hashing and EUF-CMA signatures).We present a generic hash-and-sign construction using a collision-resistant hash function and a standard signature scheme, and show how to integrate it with hash-chained logs, Merkle-tree anchoring, and trusted execution environments. We implement a prototype library and report microbenchmarks on commodity hardware, demonstrating that per-event overhead is small and predictable. This work aims to provide a foundation for standardized audit mechanisms in regulated AI, with implications for clinical trial management, pharmaceutical compliance, and medical AI governance.

Network Intrusion Detection Systems (NIDS) are essential tools for detecting network attacks and intrusions. While extensive research has explored the use of supervised Machine Learning for attack detection and characterisation, these methods require accurately labelled datasets, which are very costly to obtain. Moreover, existing public datasets have limited and/or outdated attacks, and many of them suffer from mislabelled data. To reduce the reliance on labelled data, we propose AutoGraphAD, a novel unsupervised anomaly detection approach based on a Heterogeneous Variational Graph Autoencoder. AutoGraphAD operates on heterogeneous graphs, made from connection and IP nodes that represent network activity. The model is trained using unsupervised and contrastive learning, without relying on any labelled data. The model's losses are then weighted and combined in an anomaly score used for anomaly detection. Overall, AutoGraphAD yields the same, and in some cases better, results than Anomal-E, but without requiring costly downstream anomaly detectors. As a result, AutoGraphAD achieves around 1.18 orders of magnitude faster training and 1.03 orders of magnitude faster inference, which represents a significant advantage for operational deployment.

During the past few years, we have witnessed various efforts to provide confidentiality and integrity for applications running in untrusted environments such as public clouds. In most of these approaches, hardware extensions such as Intel SGX, TDX, AMD SEV, etc., are leveraged to provide encryption and integrity protection on process or VM level. Although all of these approaches increase the trust in the application at runtime, an often overlooked aspect is the integrity and confidentiality protection at build time, which is equally important as maliciously injected code during compilation can compromise the entire application and system. In this paper, we present Tical, a practical framework for trusted compilation that provides integrity protection and confidentiality in build pipelines from source code to the final executable. Our approach harnesses TEEs as runtime protection but enriches TEEs with file system shielding and an immutable audit log with version history to provide accountability. This way, we can ensure that the compiler chain can only access trusted files and intermediate output, such as object files produced by trusted processes. Our evaluation using micro- and macro-benchmarks shows that Tical can protect the confidentiality and integrity of whole CI/CD pipelines with an acceptable performance overhead.