SECRYPT 2026 Abstracts

Area 1 - Applied Cryptography And Network Security

Full Papers

Paper Nr:	42
Title:	Design and Performance Evaluation of Post-Quantum Authentication for Embedded Systems: A Case Study on PIV
Authors:	Emmanuelle Dottax and Rina Zeitoun
Abstract:	As the transition to post-quantum cryptography accelerates, security protocols must evolve to resist quantum threats while remaining practical, particularly on constrained devices where memory, bandwidth, and performance are limited. We consider the NIST Personal Identity Verification (PIV) system, where smart cards rely on digital signatures for authentication. Since post-quantum signatures introduce substantial computational and memory overhead, whereas post-quantum Key Encapsulation Mechanisms (KEMs) are generally lighter, we investigate KEM-based alternatives for authentication and assess the migration of secure messaging to post-quantum primitives. We propose post-quantum variants of the PIV authentication and secure messaging protocols and implement both signature-based and KEM-based approaches on a real smart card platform. We evaluate their computational and communication costs in a realistic embedded setting and present detailed performance metrics that enable assessing the impact of post-quantum migration across different hardware and communication configurations. Our results show that KEM-based authentication significantly reduces execution time and transmitted data compared to post-quantum signature-based designs, while KEM-based post-quantum secure messaging incurs moderate overhead compared to its classical counterpart. These findings highlight KEM-based authentication as a practical migration strategy for post-quantum secure embedded systems.
Download

Paper Nr:	58
Title:	ISI: A New Tool for Instruction Integrity against Fault Injection Attacks Using Low-Latency MAC
Authors:	Fatimaezzahraa Boutagouaouin, Emmanuel Amankwah, Vianney Lapôtre, Hélène Le Bouder and Gaël Thomas
Abstract:	This paper presents a proof of concept for a new approach to ensuring the integrity of instructions against fault injection attack (FIA). This countermeasure, called instruction set integrity (ISI), is a cryptographic solution. The main idea is to compute with low latency a tag of each instruction using a message authentication code (MAC). The ISI ensures both the integrity of individual instructions and compliance with the control-flow graph (CFG). This solution solves the problem of multi-predecessor for an instruction in a program.
Download

Paper Nr:	59
Title:	COBRA: A Content-Agnostic Framework for Zero-Day Detection of Suspicious Domains
Authors:	Alexandros Fourtounis, Emmanouil Papadogiannakis, Panagiotis Papadopoulos, Nicolas Kourtellis and Evangelos Markatos
Abstract:	The use of malicious domains is central to cyberattacks such as phishing, malware distribution, impersonation, and fraudulent transactions. Because domains are inexpensive to register and easy to deploy at scale, they remain one of the most common and damaging tools used in cybercrime across industries. Proactive detection is essential to reducing this window of vulnerability and preventing harm to users. In this work, we propose COBRA: a content-agnostic, registration-time detection framework for identifying and analyzing suspicious domains from day zero. Our approach does not rely on any content-based features, allowing us to classify a domain even before it is populated with content. We analyze the names of newly registered domains and employ a clustering technique to group them based on lexical and structural similarity. We evaluate our methodology using real-world data consisting of 1.5M newly created domains, demonstrating that COBRA detects suspicious domains with a precision of 98.5%, identifying more than 47K distinct newly registered suspicious domains. Furthermore, our results show that domain-name clustering enables accurate early detection, allowing us to identify 80% of suspicious or malicious domains earlier than one of the most widely used threat-intelligence services, which in some cases may require up to 7 days.
Download

Paper Nr:	74
Title:	Unveiling Hierarchical Machine Learning UDP–QUIC Intrusion Detection: Protocol-Aware Flow Analysis and a New Cloud-Generated DDoS Dataset
Authors:	Sepehr Jafari, MohammadMoein Shafi and Arash Habibi Lashkari
Abstract:	The increasing adoption of UDP and QUIC for latency-sensitive and encrypted communication has created new challenges for intrusion detection, as traditional payload-centric and signature-based approaches offer limited visibility into these protocols. This paper introduces a hierarchical intrusion detection framework that operates entirely on flow-level features and combines a fast UDP-only classifier with a higher-capacity UDP+QUIC decision layer to resolve uncertain cases. To support rigorous evaluation, we also present a new cloud-generated dataset that includes diverse benign enterprise activities, multiple UDP-based DDoS campaigns, and protocol-aware flow representations extracted with our UDPFlowLyzer and QUICFlowLyzer analyzers. Experimental analysis conducted using the newly generated BCCC-UDP-QUIC-IDS-2025 dataset demonstrates that the proposed architecture improves detection reliability, enhances discrimination among attack types, and benefits from integrating QUIC-derived features, particularly for flows that are inherently ambiguous at the UDP layer alone. The results confirm the effectiveness of the hierarchical approach and highlight the value of standardized, protocol-aware feature extraction for intrusion detection in encrypted and connectionless traffic environments
Download

Paper Nr:	82
Title:	CT-Monitor++: Improving the Certificate Transparency Ecosystem with Cryptographic and Security Checks
Authors:	Simone Orlando, Alessandro Barenghi and Gerardo Pelosi
Abstract:	Digital certificates are one of the main tools to provide data and public key authentication, but their use in a Public Key Infrastructure (PKI) relies on the trustworthiness of each individual Certification Authority (CA), which may be compromised or act maliciously. Certificate Transparency (CT) mitigates this risk by requiring all issued certificates to be publicly logged, enabling the detection of unauthorized certificates. In this work, we present the design of a CT monitor that goes beyond traditional domain-based tracking by performing systematic security analysis on all certificates recorded in public CT logs. Our monitor executes a comprehensive set of security checks, inspecting certificates for cryptographic weaknesses, syntactic inconsistencies, and potentially malicious contents, providing real-time insights into the security of the Web PKI ecosystem. We implement and evaluate our monitor on live CT log streams, analyzing its computational overhead, memory requirements, and scalability as certificate volumes increase. We quantify the resources needed to perform these analyses in real time and show that, with the proposed optimizations, our monitor is capable of keeping up with the current certificate issuance rates while performing all security checks.
Download

Paper Nr:	89
Title:	Ringslip: Ring Signatures from the Lattice Isomorphism Problem
Authors:	Callum London, Daniel Gardham and Constantin Cătălin Drăgan
Abstract:	Ring signatures provide authentication over messages, whilst providing anonymity amongst a set of signerdefined public keys. They see active use in cryptocurrencies, e-voting and concurrent signature domains. However, post-quantum constructions typically rely on lattices, specifically utilising the Learning with Errors (LWE) and Short-Integer-Solution (SIS) problems, which cause inefficiencies when compared with classical constructions. One promising route to circumvent the inherent challenges of these underlying assumptions is the Lattice Isomorphism Problem (LIP), which underpins the HAWK signature scheme by Ducas et. al, currently a second round candidate in the NIST standardisation project Post Quantum Cryptography: Additional Digital Signature Schemes. It offers significant performance improvements over standard lattice assumptions due to its improved decoding, however, the only known construction of a ring signature from LIP has been shown to not satisfy linkability nor correctness. In this paper we propose RingSLIP, a secure linkable ring signature based on LIP, utilising the HAWK signature. The resulting ring signature is logarithmic in the number of ring members, and concretely has size 46KB when targeting 128 bits of security for 4096 ring members, which is competitive with other latticebased schemes. Furthermore, we observe that our construction also benefits from online/offline computation, resulting in a signature with online signing and verification only requiring 8.54×10^4 and 1.48×10^5 CPU cycles respectively, compared to 1.35×10^11 without these optimisations.
Download

Paper Nr:	94
Title:	Scalable High-Throughput FPGA Architecture for SMAC Message Authentication Code
Authors:	Ahmet Malal, Hakan Güler, Bahadır Aydoğan and Oğuz Yayla
Abstract:	SMAC is a recently proposed by Wang et al. stand-alone Message Authentication Code (MAC) constructed from repeated applications of the AES round function and featuring an aggregation mode, SMAC-1×n, for scalable parallel processing. Although originally designed for high-throughput CPU implementations leveraging AES-NI instructions, its structural properties suggest strong compatibility with hardware parallelism. However, no systematic FPGA-oriented architectural study of SMAC has been reported. This paper presents a scalable FPGA architecture of SMAC implemented on a Xilinx Kintex UltraScale+ KCU116 platform. The Π transformation is evaluated in a single clock cycle using fully combinational AES rounds, and throughput scaling is achieved through physical replication of aggregation lanes. All SMAC-1×n configurations up to n = 16 are implemented and evaluated. Post-implementation results achieve maximum operating frequencies up to 526 MHz and peak throughput of 731 Gbps for SMAC-1×16. The design exhibits near-linear throughput scaling up to eight lanes and reaches a maximum efficiency of 24.5 Mbps/slice. These results demonstrate that SMAC’s round-based construction is well suited for FPGA parallelism and enables competitive highthroughput hardware MAC acceleration.
Download

Paper Nr:	96
Title:	HomeGuard: Community-Driven Hierarchical Federated Learning for Robust Smart-Home Intrusion Detection
Authors:	Philipp Eichhammer, Christian Berger and Hans P. Reiser
Abstract:	Federated Learning (FL) has emerged as a promising approach to build collaborative Intrusion Detection Systems (IDSs) in the IoT, e.g., in smart homes. FL allows models to be shared without exposing sensitive training data, thus protecting the privacy of IoT users. However, existing FL-based IDSs rely on assumptions that rarely hold in practice, namely homogeneous devices, synchronous participation, and benign contributors. We argue that, in real-world smart homes, IoT devices are highly heterogeneous, resource-constrained, and attractive targets for adversaries, which makes conventional FL less effective or vulnerable to poisoning attacks. We present HOMEGUARD, a collaborative IDS specifically designed for the constraints and threat model of practical smart home IoT infrastructures. In our approach, we rethink FL deployment by (1) offloading model training to gateways to manage computational heterogeneity of IoT devices and (2) organizing anomaly detection models into device-specific communities based on privacy-preserving traffic fingerprints which do not expose sensitive data. Within communities and across smart homes, HOMEGUARD implements an asynchronous, hierarchical FL architecture that tolerates device churn, uneven data availability, and Byzantine participants. Further, HOMEGUARD applies Byzantine-robust aggregation at two levels: within local communities, and globally in the cloud to limit the impact of compromised devices. Experimental evaluation shows that HOMEGUARD achieves an average true positive rate of 97.86% locally and 97.53% globally with a 0% false positive rate, while maintaining robustness against both targeted and untargeted poisoning attacks.
Download

Paper Nr:	101
Title:	A Framework for Context-Aware Read Authorization over Encrypted Data
Authors:	Roberta Cimorelli Belfiore, Anna Lisa Ferrara and Barbara Masucci
Abstract:	Many security and privacy-sensitive systems must control access to encrypted data based on dynamic operational context, such as workflow stage, safety mode, or emergency conditions. Data producers cannot anticipate which context will apply at processing time, and the component observing context cannot be trusted with plaintext or policy structure. Existing mechanisms, including Access Control Encryption (ACE), assume that all authorization-relevant information is fixed at encryption time, and offer no mechanism to incorporate context that becomes available only after the data has been produced and forwarded. We introduce Contextual Reading ACE (CR-ACE), a framework for enforcing context-dependent read authorization without a trusted reference monitor. An honest-but-curious intermediary sanitizes sender ciphertexts and attaches public contextual attributes while remaining oblivious to message contents, principal identities, and authorization logic. Receivers locally enforce authorization as the conjunction of the global ACE policy and a contextual predicate embedded in their key material. We formalize a security model for CR-ACE, provide a construction and prove that it achieves contextual payload privacy, sender anonymity, sanitization security under context, and correct enforcement of both global and context-dependent authorization in our non-collusion model.
Download

Paper Nr:	116
Title:	Finding Fake Base Stations with Directional Antenna Using Measurement Reports
Authors:	Tuomo Lehtilä, Sanish Gurung, Mohamed Taoufiq Damir, Amy Sokhna Sidibé, Gizem Akman and Valtteri Niemi
Abstract:	Fake base stations (FBSs) can carry out attacks against mobile devices and user equipment (UE). We tackle the problem of localizing an FBS after its existence has been detected. Karaçay et al. (2021) have presented a localization method based on the values of the Reference Signal Received Power (RSRP), reported in the standardized measurement reports of the 3rd Generation Partnership Project (3GPP) and sent by the UEs. We show that the method they have proposed fails if the FBS uses a directional antenna (instead of an omnidirectional antenna). The localization method by Karaçay et al. is based on half-planes. We propose a new method that localizes FBSs using a directional antenna by first employing a half-plane method and then applying a method based on convex hulls. Our tests for the new method on data derived from Network Simulator 3 (ns-3) show that it significantly outperforms the previous method for a directional FBS antenna.
Download

Paper Nr:	141
Title:	Threat-Reactive Encryption: Real-Time ML-Driven Key Rotation for Adaptive Cryptographic Defense
Authors:	Mohammed Elhajj
Abstract:	Static key rotation policies are structurally mismatched to machine-paced adversaries that can exfiltrate session keys within a single cryptoperiod. This paper presents Threat-Reactive Encryption (TRE), a four-layer framework that couples a real-time ML threat scorer to a PKCS#11-compatible key management pipeline, closing the gap between threat detection and cryptographic response. A 47-feature telemetry vector from network flows, TLS handshake metadata, syscall traces, and hardware performance counters (HPCs) is scored every 500 ms by an XGBoost + Isolation Forest ensemble inside a hardware-attested Trusted Execution Environment (TEE). Rotation fires atomically when the unified threat score exceeds a Bayesian adaptive threshold, installing new key material via TLS 1.3 post-handshake authentication without session interruption. On a 168-hour, 10 Gbps testbed with 500 concurrent TLS 1.3 sessions and 12 injected CALDERA attack campaigns, TRE achieves macro-AUPRC = 0.952, median rotation latency = 57 ms (P99 = 163 ms), and steadystate FPR≤2.0% with zero session resets. Post-compromise key exposure is reduced 625× vs. the 24-hour static policy and 26× vs. the 1-hour static policy. Adversarial analysis shows full evasion requires the attacker to operate at ≈7% of unconstrained attack throughput. The approach is PKCS#11-compatible and provides a direct extension path to post-quantum algorithm substitution via CRYSTALS-Kyber (NIST FIPS 203).
Download

Paper Nr:	142
Title:	Slice & Dice: Privacy-Preserving Layered Attestation via Active Memory Introspection
Authors:	Nikolaos Varvitsiotis, Stefanos Vasileiadis, Sofia Anna Menesidou, Thrasyvoulos Iliadis, Konstantinos Nikas, Nectarios Koziris and Thanassis Giannetsos
Abstract:	The rapid adoption of dynamic Payload Managers enables portable, platform-agnostic execution across the cloud/edge continuum. However, these complex execution layers, even when embedded within a Trusted Execution Environment (TEE), significantly expand the attack surface of the Trusted Computing Base (TCB). Existing layered attestation schemes typically combat this using static, load-time measurements of payloads. Yet, recent vulnerabilities demonstrate that static checks are insufficient to prevent attackers from gaining privileged execution rights while avoiding detection. Slice and Dice proposes a layered attestation architecture for 3-layer systems, in which a minimal and highly privileged programmable firmware layer (Layer 0) continuously performs dynamic memory introspection on the untrusted Payload Manager layer (Layer 1). By inspecting ELF segments during execution, the firmware can detect code modifications in real-time. To complete the cycle of Remote Attestation (RA) and provide evidence to an external verifier in a privacy-preserving manner, we couple the introspection mechanism with Zero-Knowledge Proofs (ZKP) utilizing BBS anonymous credentials.
Download

Paper Nr:	145
Title:	PQKE-hCPABE: A Hybrid CP-ABE Scheme with Post‑Quantum Key Exchange for Secure Multicast
Authors:	Matteo Sandrucci, Rana Abu Bakar, Abraham Cano Aguilera, Francesco Fumagalli, Francesco Paolucci, Juan Jose Vegas Olmos, Piero Castoldi and Filippo Cugini
Abstract:	In this paper, we present a SmartNIC/DPU-accelerated secure multicast framework that integrates Ciphertext Policy Attribute Based Encryption (CP-ABE), post quantum key encapsulation (ML-KEM / ML-KEM-768), and AES-256-GCM data-plane encryption into a unified architecture. The design decouples control-plane security operations from the data plane, enabling efficient session key distribution via CP-ABE and ML-KEM, while offloading per-packet encryption and decryption to hardware using DPUs and the DOCA framework. We implement a fully functional both unicast and multicast transmission modes, where a single encrypted stream is securely distributed to multiple receivers. Each receiver independently reconstructs the session key using attribute-based policies and post-quantum decapsulation, ensuring fine-grained access control and quantum-resilient security for key exchange. The data plane leverages hardware-accelerated AES-GCM-256 with asynchronous task execution, zero-copy memory, and batched packet processing to achieve high throughput.
Download

Paper Nr:	159
Title:	A Capability-Driven Framework for Prioritizing Post-Quantum Cryptography Migration in Enterprise Architectures
Authors:	Francis Wanko Naa, Nora Boulahia-Cuppens and Frederic Cuppens
Abstract:	Post-quantum cryptography (PQC) migration introduces a critical prioritization challenge: organizations must determine which cryptographic mechanisms to transition first within complex and interdependent systems. Existing approaches largely focus on algorithmic vulnerability and provide limited support for system-level decision-making. This paper proposes a capability-driven framework for PQC prioritization that integrates architectural context and cyber-resilience considerations. The approach combines a structured cryptographic registry with a multi-criteria scoring model capturing dependency centrality, business criticality, post-quantum exposure, propagation impact, and migration complexity. Beyond cryptographic risk, the proposed approach explicitly supports cyber-resilience objectives by identifying mechanisms whose compromise would have the greatest impacts on enterprise capabilities. The framework is applied to a real-world enterprise system, enabling a systematic evaluation of cryptographic mechanisms across multiple services and architectural layers. The results show that prioritization is not solely driven by cryptographic vulnerability but emerges from the interaction between quantum exposure and architectural positioning. In particular, trust-establishing mechanisms-such as key exchange and identity-related functions-consistently rank highest due to their transversal role and system-wide impact. These findings highlight the limitations of vulnerability-driven strategies and demonstrate how the proposed framework supports traceable, context-aware, and iterative migration planning to align PQC migration with enterprise architecture. From a cyber-resilience perspective, the results show how prioritization supports the ability of systems to anticipate, withstand, and adapt to quantum-induced disruptions.
Download

Paper Nr:	182
Title:	A Maliciously Secure and Fully Decentralized Threshold FHE Scheme with Native RNS Acceleration
Authors:	Ting-Yu Chen, Arijit Karati, Er-Shuo Zhuang and Chun-I. Fan
Abstract:	Threshold fully homomorphic (ThFHE) encryption, as a communication encryption protocol, ensures that no third party participates in generating or knows any parameters. Compared to multi-key fully homomorphic encryption, it avoids excessive noise expansion caused by too many users participating in the calculation. However, current ThFHE algorithms focus on reducing computational overhead, thereby neglecting integrity verification of participating nodes’ behavior in distributed collaborative environments, leaving the system vulnerable to malicious actors. Without an effective verification mechanism, malicious nodes can manipulate the final result without breaking the protocol flow by injecting biased noise or providing forged partial decryption values, compromising data integrity. This research proposes an enhanced ThFHE encryption scheme based on a Full Remainder System (Full-RNS) architecture. This scheme integrates Distributed Key Generation (DKG) and Multi-Party Computation Relinearized Key (MPC RLK) techniques to achieve fully decentralized parameter initialization. To combat malicious attacks, we introduce a Non-Interactive Zero-Knowledge (NIZK) proof that incorporates smudging noise, ensuring that the computational trajectory at each stage can be publicly verified without leaking private key information. The results of the experiment show that this scheme maintains efficient homomorphic computation of the BFV algorithm while effectively resisting node fraud, providing more robust security for voting systems and medical privacy computations.
Download

Paper Nr:	194
Title:	FLiPD: Privacy-Preserving Federated Learning via Multi-Party Computation and Differential Privacy
Authors:	Gowri R. Chandran, Melek Önen and Thomas Schneider
Abstract:	Federated Learning (FL) is a collaborative Machine Learning (ML) process where clients locally train an ML model on their private inputs, and send it to a server that aggregates these to obtain a global model update. FL is widely used in applications where the training data is distributed among several clients, e.g., for next word prediction in Gboard. Nevertheless, FL faces several privacy and security challenges. 1) Client privacy needs to be preserved by employing defenses against inference attacks using Secure Aggregation (SA) protocols. 2) The security of the model has to be defended against poisoning and backdoor attacks, e.g., by using clustering or filtering algorithms. In this work, we present FLiPD, an optimised SA protocol for FL that protects against several attacks via a combination of Multi-Party Computation (MPC) and Differential Privacy (DP) mechanisms. We provide defenses against both inference and backdoor attacks. As opposed to existing solutions, in FLiPD, the clientserver communication cost is essentially the same as in unprotected FL, which sends plaintext updates. Furthermore, the server-server communication cost is slightly lower (by 11%) than the state-of-the-art Prio+ (Addanki et al., SCN’22). In addition, we examine the accuracy of FLiPD both in the presence and absence of attacks. We achieve 87% accuracy for a Linear Regression model trained on the HAR dataset, and 90% for a Convolution Neural Network trained on the MNIST dataset.
Download

Paper Nr:	195
Title:	On Automating the Obfuscation of Malware Payloads: Zone, a bin2bin Metamorphic Engine
Authors:	Stefano Bistarelli, Jacopo Di Benedetto and Francesco Santini
Abstract:	We developed Zone, an open source metamorphic engine that employs a full bin2bin approach to obfuscate small payloads. We also developed a novel metamorphic technique that evades detection by shuffling instructions while preserving the original execution flow. We extensively evaluated our engine on a dataset of 102 payloads for Windows x64 and x86 architectures. The tests we performed on VirusTotal showed a significant reduction in detection rate compared with identical payloads without metamorphic transformations. Finally, we compared our engine against widely adopted payload obfuscators, such as Shikata Ga Nai (x86/x64). Our results show that our automatic strategy can match or outperform traditional encoders in static detection while avoiding runtime exposure of the original payload.
Download

Paper Nr:	223
Title:	Experimental Evaluation of Lightweight Encryption Algorithms on 16-bit Microcontrollers
Authors:	Marino Miculan, Matteo Paier and Jacopo Plozner
Abstract:	The Internet of Things (IoT) increasingly relies on resource-constrained devices that require efficient symmetric encryption. Lightweight cryptography (LWC) addresses this by providing algorithms optimized for minimal computational and energy overhead. However, the performance of any cryptographic primitive is highly architecture-dependent, and direct experimental data on microcontrollers remains scarce in the literature. In this paper we present a rigorous evaluation of ten lightweight encryption algorithms - seven block ciphers (AES, CLEFIA, LEA, PRINCE, QARMAv2, SPARX, SPECK) and three stream ciphers (Ascon, ChaCha20, Hummingbird-2) - running on the 16-bit Texas Instruments MSP430FR6989 microcontroller, a platform widely used in industrial IoT and smart metering. Unlike prior surveys that rely solely on software counters or emulation, our methodology employs professional-grade instrumentation for real-time current profiling at the hardware level. Each algorithm is characterised under two compiler optimisation profiles (speed-optimised and size-optimised) across three metrics: encryption throughput, code footprint, and charge consumption. Our results show that ARX-based ciphers - particularly LEA and SPECK 32/64 - achieve the best energy efficiency, outperforming AES by up to 25% in charge consumption while maintaining a significantly smaller code footprint. Hardware-oriented designs (PRINCE, QARMAv2) perform poorly in software, confirming that hardware efficiency does not translate to software performance. Among stream ciphers, Ascon - the NIST LWC standard - offers the best balance of security and efficiency, whereas ChaCha20 proves unsuitable for heavily resource-constrained contexts. We provide concrete algorithm recommendations for developers targeting MSP430 and similar 16-bit platforms, and we openly release all benchmark implementations.
Download

Paper Nr:	229
Title:	A PQC-Enabled IoT Trust Architecture
Authors:	Davide Ferraris, Kun Sun, Sushil Jajodia and Javier Lopez
Abstract:	The rapid growth of Internet of Things (IoT) deployments introduces significant security challenges, particularly in the context of device authentication, trust management, and anomaly detection. At the same time, the advent of quantum computing threatens the long-term security of classical cryptographic mechanisms commonly used in IoT ecosystems. In this paper, we present a novel IoT trust framework that integrates post-quantum cryptography (PQC), a real-time anomaly detection, and distributed auditability. Devices are onboarded using PQC primitives, combining ML-DSA-based signatures for authentication and ML-KEM for secure session establishment. Runtime behaviour is continuously monitored through telemetry streams, which are analyzed by a light-ai monitor in order to detect anomalous activity using a sliding-window trigger mechanism. Security-relevant events are recorded in an append-only ledger, ensuring traceability and auditability of system decisions. We evaluate the proposed architecture through experimental runs in a containerized environment. Results show that PQC-based onboarding introduces minimal latency overhead, while the monitoring system effectively detects anomalous behaviour. The system demonstrates a clear separation between benign and anomalous device behaviour, with consistent trigger values exceeding detection thresholds. The proposed framework provides a practical and forward-looking solution for securing IoT systems against both classical and quantum-era threats, combining strong cryptographic guarantees with adaptive behavioural monitoring.
Download

Short Papers

Paper Nr:	43
Title:	Third-Party Script Identification through Supervised Contrastive Learning on Behavioral Profiles
Authors:	Pedro Carmo Pereira, João Routar, Susana Lima, Jorge Meira and Pedro Fortuna
Abstract:	Web applications increasingly rely on third-party scripts for critical functionality, creating complex chains of unmonitored dependencies that complicate compliance with PCI DSS v4.0 Requirement 6.4.3. This standard mandates an up-to-date inventory of all scripts on payment pages, a task impractical for manual verification due to the dynamic nature of modern web ecosystems. This paper presents a supervised contrastive learning framework to automate third-party script inventorying based on runtime behavioral profiles. Our methodology learns embeddings of behavioral sequences to recognize known vendors and detect previously unseen scripts. Evaluated on a dataset from 18,000 web applications, our approach achieves 81.46% accuracy in identifying known vendors and an 83.03% Unknown Detection Rate (UDR), significantly outperforming traditional classifiers. Furthermore, the model can recognize distinct but previously unseen vendors without retraining, achieving an 87.91% Within-Unknown Identification Rate. This provides a scalable solution for the evolving JavaScript ecosystem, directly supporting automated compliance and security monitoring.
Download

Paper Nr:	48
Title:	Are Explanations Robust to Federated Poisoning? An Empirical Study in Cyber Threat Detection
Authors:	Daniel Bakhtiari, Mina Alishahi and Gaurav Choudhary
Abstract:	Federated learning has emerged as a promising approach for cybersecurity applications by enabling collaborative model training without sharing sensitive data. At the same time, explainable artificial intelligence (XAI) is increasingly used to provide transparency in artificial intelligence models. However, little is known about how adversarial poisoning attacks in federated learning affect the reliability of explanations produced by XAI. This paper investigates the impact of adversarial poisoning attacks on explanation quality in federated learning. An empirical evaluation is conducted using three benchmark cybersecurity datasets and multiple adversarial poisoning scenarios, including label-flipping and model-poisoning attacks with different levels of severity. The experiments are evaluated using XAI methods: Integrated Gradients (IG), SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME).
Download

Paper Nr:	61
Title:	A Game-Theoretic Approach for Optimal Multi-Target Defense Strategies in Programmable Networking
Authors:	Jamil Ahmad Kassem, Helena Rifà-Pous and Joaquin Garcia-Alfaro
Abstract:	Traditional network defense strategies, which follow a linear sequence of vulnerability discovery, defense selection, and attack mitigation, often struggle to adapt to emerging and unpredictable cyber threats. This paper presents a novel strategic framework designed to optimize defense costs, addressing both security concerns and cost-effectiveness. Drawing inspiration from security games and Stackelberg-style leader–follower interactions, our approach introduces a resource management strategy that optimizes a defender-side cost function under rational attacker behavior. We test our approach within the programmable networking paradigm, which is expected to expand traditional network architectures. Our numerical evaluation in SDN-like scenarios shows that the proposed strategy significantly reduces total defense costs compared to representative Moving Target Defense (MTD) baselines, while providing defenders with the flexibility to trade off attack impact against reconfiguration costs.
Download

Paper Nr:	63
Title:	Compact and Low-Power ECC Processor Implementation over GF(2^m) for Resource-Constrained IoT Devices
Authors:	Salah Harb, Tariq Daradkeh and Yanal Alahmad
Abstract:	Elliptic Curve Cryptography (ECC) is widely adopted for securing Internet-of-Things (IoT) devices because it delivers equivalent security to RSA at a fraction of the key size, yet resource-constrained devices demand implementations that minimize silicon area without relying on embedded hard macros. This paper presents a compact, fully parameterized FPGA cryptoprocessor for scalar multiplication over the NIST-standardized Koblitz curves defined over GF(2m) for m ∈ {163, 233, 571}, covering security levels from 80 to 256 bits. The design exploits the Frobenius endomorphism with τ-NAF scalar recoding to replace all point doublings with cost-free squarings, employs a bit-serial polynomial-basis multiplier, and introduces a novel single shared operand bus microarchitecture that collapses three independent multiplexer trees into one centralized m-bit bus with a two-phase loading protocol. The architecture is parameterized by field size, curve coefficient, and irreducible polynomial, sharing an identical 41-state FSM across all field instances. Implemented on a low-cost Artix-7 FPGA using Xilinx Vivado 2024.2, the proposed design achieves as few as 1,386 slices for m=163 at 140 MHz—the lowest slice count among comparable logic-only implementations on 7-Series devices—and scales efficiently to m=233 and m=571, all without any DSP blocks or Block RAM. The FPGA results demonstrate that the proposed architecture is well suited for resource-constrained IoT devices across the full range of NIST-recommended Koblitz binary fields.
Download

Paper Nr:	97
Title:	A Bridged Sandbox–Honeypot Architecture for Behavioral Analysis of ICS Malware
Authors:	Dawit Berhan, Stefano Bistarelli, Benedetto Manoni and Francesco Santini
Abstract:	The convergence of IT and OT networks has exposed Industrial Control Systems (ICS) to targeted malware that abuses industrial protocols. Standalone sandboxes lack the industrial context required to activate OT-specific payloads, whereas honeypots alone cannot safely execute untrusted binaries. This paper presents an integrated architecture that combines a Cuckoo3 sandbox with a programmable high-interaction honeypot (Honeybus, an extension of LOGistICS, a medium-interaction OT honeypot for Modbus and S7comm) and a forensic visualization tool (ViewMod). We validate the architecture by executing FrostyGoop, a Modbus-based ICS malware, against a simulated water-supply control system. The current evaluation focuses on Modbus and a single malware family. Performance benchmarking, systematic evasion testing, and comparison with other honeypots are explicitly identified as future work.
Download

Paper Nr:	104
Title:	Privacy-Preserving Counterfactual Explanations for Federated AI
Authors:	Sjoerd Berning, Vincent Dunning, Thijs Veugen and Kevin Witlox
Abstract:	As the usage of Artificial Intelligence (AI) for sensitive purposes increases, there is a growing need for privacy-aware explainable AI (XAI) tools. In this paper, we present a privacy-preserving counterfactual explanation algorithm. Our starting point is a decision-support model that is able to operate on vertically partitioned datasets, meaning that each party holds a different subset of datapoint attributes. The goal of a counterfactual algorithm is to find, given an observation, a datapoint from the (virtual) dataset that is closest to the observation but has a different label. Our algorithm fully preserves the privacy of the n datapoints belonging to the different parties by combining the strengths of homomorphic encryption and secret sharing. Through a number of experiments, we demonstrate the added value of combining multiple datasets in a realistic scenario and show that the privacy-preserving solution does not affect the accuracy. We fully implement our solution and demonstrate that it scales as to thousands of datapoints.
Download

Paper Nr:	127
Title:	Optimizing SPN Full Key Recovery via MILP-Aided Trail Selection
Authors:	Abhilash Kumar Das and Aditi Gautam
Abstract:	In this paper, we present an efficient Mixed Integer Linear Programming (MILP)-aided key recovery framework that enables the systematic selection of differential trails with minimized subkey enumeration cost. Our approach formulates trail selection as an optimization problem, allowing the cryptanalyst to balance statistical strength and computational complexity in a principled manner. To improve key recovery results, we optimize trail selection to decrease enumeration costs while maximizing the probability of a correct guess. The proposed technique is generic and applicable to a wide range of symmetric-key primitives, particularly SPN-based block ciphers. As a case study, we demonstrate a full key recovery attack on the FUTURE block cipher using our optimized multi-trail strategy. Additionally, we report a reduced number of Work Units (WUs) in the MILP-based trail search compared to previous approaches, highlighting the practical efficiency of our method.
Download

Paper Nr:	147
Title:	PEACE: Privacy-Enhanced Authentication for Cryptocurrency Environments
Authors:	Stefan Dziembowski, Shahriar Ebrahimi, Paweł K˛edzior, Michał Król and Tomasz Lizurej
Abstract:	As blockchains gain traction in managing financial assets, robust authentication mechanisms become essential. Traditional public-private key authentication poses risks of key mismanagement, leading to significant financial losses. At the same time, in the non-blockchain context, systems like OpenID Connect (OIDC) provide user-friendly authentication with identity recovery in the traditional Web. However, they rely on a single trusted party, raising privacy concerns when applied to blockchains. In this paper, we present PEACE, an authentication mechanism for blockchains that combines the practicality of traditional Web systems with decentralization and privacy-preservation, building on a recent groundbreaking zkLogin protocol of Baldimtsi et al. (ACM CCS 2024). The protocol frees users from managing cryptographic material without giving away fund custody or relying on a single party. We detail PEACE’s design, prove its properties in the Universal Composability framework, and present a real-world implementation compatible with existing OIDC deployments and the Ethereum blockchain.
Download

Paper Nr:	149
Title:	A Partitioned Neural Network Architecture for Efficient Inference with Fully Homomorphic Encryption
Authors:	Shusaku Uemura and Kazuhide Fukushima
Abstract:	Fully homomorphic encryption (FHE) enables computations on ciphertexts without decryption. This property is expected to be utilized in AI with sensitive data. Although encryption improves the security of neural network inference, it incurs a significant computational overhead because all processes are executed under encryption. However, in many practical scenarios, not necessarily all the input features should be encrypted. Some features must be confidential whereas others can be disclosed to the model operator. Based on this observation, we propose a novel neural network architecture which partitions the input features into two types according to their secrecy. Our architecture decomposes a neural network into three modules to handle these two feature types efficiently. Public input features are processed without encryption whereas private input features are computed under encryption. We theoretically analyze the computational cost of our model and formulate the reduction rate in terms of the parameters. We also experimentally examine our model’s accuracy by comparing it to that of a standard model and demonstrate that our model reduces the computational costs by 50% for a certain parameter set while the accuracy degradation is limited.
Download

Paper Nr:	154
Title:	ADASEC-TB: Cryptography Verification & Benchmarking for Next-Generation Networking Hardware
Authors:	Amina Tankovic and Enio Kaljic
Abstract:	Next-generation networks aim to provide a wide range of services which require high level of performance and security. In some services, security is particularly important and lack of it can lead to critical consequences due to sensitivity of the information being transmitted. Therefore, research in the field of security has significantly increased and a number of security solutions for next-generation networks are already developed. For every security solution, there is a need for a trusted test environment which can be used for its functional verification and performance evaluation. Existing security testbeds do not provide reproducible testing and evaluation of hardware-based cryptography solutions, which are especially important for high-speed next-generation networks. To overcome this limitation, we present a modular, flexible testbed which comprises several tasks: traffic generation, traffic processing (i.e. performing desired network or security functions) and analysis of traffic obtained after its processing. Suitability of our testbed for hardware-based cryptography solutions is demonstrated on an example of selective encryption through RTL simulation using ModelSim simulator.
Download

Paper Nr:	174
Title:	A Flexible Zero-Knowledge Architecture for Decentralized Energy Markets
Authors:	Hamza Zarfaoui, Victor Languille and Gérard Memmi
Abstract:	Decentralized energy markets require settlement mechanisms that are publicly verifiable while preserving the confidentiality of bids, allocations, and participant activities. In prior work, we introduced the Privacy-Preserving Exchange Mechanism (PPExM) (Languille et al., 2024), a protocol for sealed-bid markets using zk-SNARKs (Groth16) (Groth, 2016) and gnark (Consensys, 2021). In this protocol, a specific arithmetic circuit, denoted as the exchange circuit, depends on the chosen market-clearing algorithm. Specifically, although this circuit includes constraints fixed by the protocol, a subset encodes the allocation function F and must be updated for each market-clearing algorithm. This component constitutes the primary bottleneck for rules involving sorting, threshold detection, or data-dependent branching, as such operations are non-trivial to implement. To address this limitation, we propose a hybrid SNARK–STARK architecture that preserves PPExM while replacing this part of the exchange circuit using a zk-STARK-based Zero-Knowledge Virtual Machine (zkVM). Now, F can simply be implemented as a Rust program executed in the RISC Zero zkVM. The resulting zk-STARK proof is then compressed into a Groth16 proof and recursively verified within the original exchange circuit. We implement the resulting system and evaluate uniform-price double auctions with up to 300 participants. The results show that separating protocol invariants from exchange computation yields a flexible and maintainable confidential-settlement framework without sacrificing succinct verification or practical relevance for smart-grid clearing.
Download

Paper Nr:	187
Title:	Area-Time Efficient Hardware Implementation of HQC Encoder with Selectable Security Levels
Authors:	Ömer Üstün and Muhammed Çelik
Abstract:	Hamming Quasi-Cyclic (HQC) is a code-based post-quantum cryptography (PQC) scheme selected by NIST for standardization as a key encapsulation mechanism (KEM). Efficient hardware implementation of HQC is important for deployment in resource-constrained devices. This paper presents an area-efficient standalone HQC encoder that supports all three HQC security levels through a unified datapath with runtime parameter selection. The architecture combines a configurable Reed-Solomon encoder with a single-cycle Reed-Muller stage and a direct stage interface that avoids intermediate buffering. Because a single unified instance replaces three separate encoder cores, the design reduces aggregate resources by up to 2.9× compared to deploying one fixed-level encoder per parameter set. On a per-level basis, experimental post-synthesis results show LUT-ADP comparable to the best prior standalone encoder at HQC-128—where the unified datapath is overprovisioned relative to a dedicated design—and up to 1.65× better at HQC-256, with improvements up to 2.55× over single-level encoder components extracted from full-KEM implementations.
Download

Paper Nr:	212
Title:	Hierarchical Temporal Evidence Fusion for Intrusion Detection in Edge–Fog–Cloud IoT Architectures
Authors:	Myria Bouhaddi and Farah Yasmine Sadok
Abstract:	The hierarchical and heterogeneous nature of Internet of Things (IoT) architectures, spanning edge, fog, and cloud layers, makes intrusion detection particularly challenging, as each layer provides only a partial view of the system. Traditional intrusion detection systems (IDS) often fail to identify coordinated and distributed attacks from fragmented observations. This paper proposes HTEF-IDS, a hierarchical intrusion detection framework that combines LSTM-based temporal modeling with Dempster–Shafer evidence fusion to infer a global security state from distributed observations. Local detections are progressively aggregated across layers, while a feedback mechanism dynamically adjusts detection sensitivity according to the inferred threat level. Experiments conducted on the CIC IoT-DIAD 2024 dataset show that the proposed framework improves detection performance, reduces false positives, and effectively identifies distributed attack patterns that are not observable at a single layer.
Download

Paper Nr:	215
Title:	Toward a Semantic-Surface Information Theory of Chain-of-Thought Steganography under Paraphrase Defense
Authors:	Ondrej Rozinek
Abstract:	Recent work on chain-of-thought (CoT) reasoning raises the concern that a language model may use its reasoning trace as a covert channel, smuggling information past an overseer who reads it only for its ostensible content. A natural defense is to paraphrase the CoT before use. The right question is then not how much total entropy CoT carries, but how much surface-form entropy survives paraphrase once semantics are fixed. We develop a stylized information-theoretic framework with three layers, a semantic representation, an honest surface realization, and a paraphrased realization, with paraphrase-robust features extracted from each. The natural hiding budget is the conditional surface entropy H(U \| S), and the recoverable bandwidth is the honest conditional mutual information I(U;V \| S). We prove a converse bounding any semantics-preserving covert encoder by this quantity, and give a conditional coding template that reduces the matching lower bound to four explicit, empirically testable assumptions. A binary-symmetric corollary recovers the Shannon expression as a sanity check. The framework yields a safety decomposition separating semantic content from discretionary surface bandwidth, and identifies the latter as the design target for paraphrase defenses. We intend this as a formal lens and a proof-of-concept validation rather than a general capacity theory or a finished measurement methodology, since the converse holds only for semantics-preserving encoders and is representation-dependent, the lower bound is conditional, and the experiments are illustrative.
Download

Paper Nr:	216
Title:	Lightweight Real-Time Wireless Intrusion Detection for IEEE 802.11 Networks on Edge Devices
Authors:	Himasha Jayasekera, Rajani Piyarathna, Chalana Ranwala, Tharusha Thilakarathna, Deemantha Siriwardana and Ayesha Wijesooriya
Abstract:	This paper addresses real-time intrusion detection in IEEE 802.11 wireless networks, where unprotected management frames and exploits such as KRACK and Kr00k pose persistent threats. Existing approaches often rely on signature-based detection, device-specific features, or synthetic oversampling, limiting their effectiveness in dynamic, resource-constrained environments. We propose a lightweight, edge-deployed Wireless Intrusion Detection System (WIDS) trained on the AWID3 dataset across eight legacy and modern attack classes. The approach combines a three-stage feature selection pipeline with cost-sensitive learning for compact 27-feature classification at low latency, and a multi-stage validation mechanism (Z-score filtering and temporal persistence) that reduces false positives in real-time. Deployed on a Raspberry Pi and evaluated with a stratified 70/30 split across five algorithms, a tuned Decision Tree achieves 99.76% accuracy and a 0.9797 macro F1-score, with 0.16 s batch inference from a model serializing under 30 KB, showing that lightweight models can deliver accurate, efficient real-time wireless intrusion detection.
Download

Paper Nr:	228
Title:	How Many Queries to Catch a Sandbagger? A Chernoff–Stein Sample-Complexity Theory of Capability Elicitation
Authors:	Ondrej Rozinek
Abstract:	Capability evaluations of frontier language models are usually reported as a yes/no judgement of whether the model hit the benchmark or sandbagged. This framing hides a quantitative question with a clean answer, namely how many queries an auditor needs. We model the auditor as running a binary hypothesis test between honest and sandbag response distributions P0,P1, so that the Chernoff–Stein lemma governs detection difficulty. At any fixed type-I budget the optimal type-II error satisfies − 1N logβ ? N (α) → Dπ, where Dπ = ∑q π(q) DKL P F 0 (· \| q) P F 1 (· \| q) is the pool-averaged conditional KL between honest and sandbag feature distributions, and the required sample size is N? (α,β) ≈ log(1/β)/Dπ. The right quantity to measure is therefore Dπ rather than the accuracy gap on a benchmark. We specialise the Stein lemma to iid elicitation on the joint query–feature variable, give a Chernoff-information corollary for symmetric errors, sketch a composite-hypothesis extension for query-conditional sandbaggers, and derive a pool-design principle that concentrates queries on high-KL items subject to a coverage budget. We then instantiate the framework on grade-school arithmetic with six commercial LLMs and four deterministic binary features at ∼360 API calls. Because we report summed per-feature marginal KLs rather than the joint-feature KL, the numbers are a separability proxy rather than the theorem quantity itself. On four models the proxy ranges from 3.98 to 5.80 nats per query and the implied N? (0.05,0.05) collapses to a single query, which reflects the crudeness of the prompt-induced sandbagger rather than the difficulty of realistic detection. The open question is how much of that gap a subtler, query-conditional adversary can close.
Download

Paper Nr:	230
Title:	Fast and Effective Syscall-Based IDS with Categories
Authors:	Diogo Bortolini, Rafael R. Obelheiro and Carlos A. Maziero
Abstract:	Host-based intrusion detection using syscalls can be hindered by high computational costs and difficulties in identifying relevant semantic and contextual patterns. The increasing complexity and volume of syscalls make intrusion detection more challenging, as traditional models struggle with high dimensionality and computational costs. Furthermore, treating syscalls in isolation limits identification of relevant semantic and contextual patterns. This work presents an anomaly detection method based on syscall categorization and grouping strategies that reduce data dimensionality while preserving predictive performance. The proposed approach replaces individual syscalls with semantically meaningful representations, organized by threat levels, functional categories, or their combination. From these representations, fixed-dimension arrays are constructed to capture different aspects of syscall behavior. The generated arrays include those that consider presence or absence of syscalls, frequency of occurrence, and relational proximity under linear and exponential scales, for individual syscalls as well as pairs of syscalls. These representations are then used to train various machine learning classifiers. Results demonstrate that the strategy significantly reduces computational cost and execution time without notable performance degradation. The best trade-off achieved yielded an F1 score of 0.99, with up to 85% reduction in runtime and 66% lower memory usage compared to the original representation.
Download

Paper Nr:	28
Title:	Security Event-Driven IP Hopping Relying on Intrusion Detection Systems with Decoupled Active Response
Authors:	Elías Grande
Abstract:	The unauthorised disclosure of an organisation’s network map poses a security risk to that organisation, as it provides potential attackers with a detailed view of its infrastructure. This risk is amplified when considering the static nature of the network. Therefore, it is necessary to develop new dynamic approaches that allow the configuration of network parameters to be modified, so that such variability can mitigate the risk when necessary. The main contribution of this work is the definition of a novel Moving Target Defence model of IP hopping activated by security events. These security events are detected by an Intrusion Detection System with decoupled active response capability which, using a DHCP server, manages to execute the hopping while maintaining the consistency of network traffic without requiring specialised network elements at the link layer. This allows the proposed solution to resolve the most important challenges in this type of scenario, such as who controls the mutation, when it is activated, and how network consistency is guaranteed.
Download

Paper Nr:	33
Title:	A Unified Evaluation of Learning-Based Similarity Techniques for Malware Detection
Authors:	Udbhav Prasad and Aniesh Chawla
Abstract:	Cryptographic digests (e.g., MD5, SHA-256) are designed to provide exact identity. Any single-bit change in the input produces a completely different hash, which is ideal for integrity verification but limits their usefulness in many real-world tasks like threat hunting, malware analysis and digital forensics, where adversaries routinely introduce minor transformations. Similarity-based techniques address this limitation by enabling approximate matching, allowing related byte sequences to produce measurably similar fingerprints. Modern enterprises manage tens of thousands of endpoints with billions of files, making the effectiveness and scalability of the proposed techniques more important than ever in security applications. Security researchers have proposed a range of approaches, including similarity digests and locality-sensitive hashes (e.g., ssdeep, sdhash, TLSH), as well as more recent machine-learning–based methods that generate embeddings from file features. However, these techniques have largely been evaluated in isolation, using disparate datasets and evaluation criteria. This paper presents a systematic comparison of learning-based classification and similarity methods using large, publicly available datasets. Each method is evaluated under a unified experimental framework with industry-accepted metrics. To our knowledge, this is the first reproducible study to benchmark these diverse learning-based similarity techniques side by side for real-world security workloads. Our results show that no single approach performs well across all dimensions; instead, each exhibits distinct trade-offs, indicating that effective malware analysis and threat-hunting platforms must combine complementary classification and similarity techniques rather than rely on a single method.
Download

Paper Nr:	38
Title:	Security Enhancements of the Quantum-Resistant Identity Authentication and Key Agreement Scheme for UAV Networks Based on Kyber Algorithm
Authors:	Mohammed Zitouni, Hicham Hameurlaine and Mustapha Bensalah
Abstract:	Secure UAV communication requires robust, quantum-resistant protocols that are resilient to physical capture and operate under severe resource constraints. This work identifies critical vulnerabilities-including desynchronization, capture attacks, and lack of forward secrecy-in the LIGKYX protocol(Xia et al., 2024). To address these flaws, we propose a novel protocol that replaces LIGKYX’s ECC with Physical Unclonable Functions (PUFs) for device authentication and integrates the post-quantum Kyber mechanism for key establishment. Enhanced with authenticated encryption and nonce-based key derivation, our protocol is formally verified (using BAN logic) to resist replay, impersonation, desynchronization, and capture attacks. Performance analysis confirms its lower computational and communication overhead, making it suitable for constrained UAV networks.
Download

Paper Nr:	56
Title:	Coppercloud: Blind Server-Supported RSA Signatures
Authors:	Nikita Snetkov, Jelizaveta Vakarjuk and Alisa Pankova
Abstract:	Privacy-preserving technologies are a well-established area of research for the modern digital systems, driven by the regulatory mandates, such as the GDPR, and growing demand for the users’ privacy. A notable example of such technologies is blind signature: a primitive that supports creation of privacy-sensitive applications, starting from secure e-voting and anonymous digital cash to privacy-preserving credentials and identity management systems. In this work, we introduce Coppercloud, a blind server-supported RSA signature scheme designed to enhance privacy in digital identity systems. Coppercloud enables a user to obtain a signature on a message, without revealing its content to the supporting server, while distributing the signing key between the user’s device and the supporting server. We formalize the security requirements for blind server-supported signing by defining an ideal functionality F BlindSpl, and prove that Coppercloud securely realizes this functionality in the Universal Composability (UC) model.
Download

Paper Nr:	81
Title:	A Hardware Implementation of Blockchain Enabled Cross Domain Authentication Scheme in Internet of Vehicles
Authors:	Priya Jyotiyana and Jayaprakash Kar
Abstract:	Internet of Vehicles (IoV) enable enable communication between vehicles and infrastructure to improve road safety and traffic efficiency. However, authenticating vehicles across different administrative domains remains a challenge. This work proposes a blockchain-based authentication framework that enables secure cross-domain communication between vehicles. The proposed architecture combines Ethereum smart contracts with a Trusted Authority (TA) and Key Generation Center (KGC) to maintain decentralized domain trust while allowing efficient local authentication. A prototype system was implemented using a WiFi Router as the Road Side Unit (RSU) for network connectivity, a Raspberry Pi 5 hosting the TA and KGC servers, ESP32 microcontrollers as vehicle nodes, and a local Ethereum blockchain using Hardhat. Experiments were conducted with vehicle counts ranging from 100 to 500 and domain counts from 2 to 10. The system maintained a packet delivery ratio of 100% across all experiments while achieving average latency between 49 ms and 206 ms and throughput between 391 and 723 requests per second. The results demonstrate that blockchain-assisted authentication can be integrated into IoV systems without significant performance degradation.
Download

Paper Nr:	86
Title:	VeiledWaves: A Covert Channel via Transmission Power Modulation in Wi-Fi Networks
Authors:	Tanja Klöffel, Tim Nöth, Tobias Fertig and Sebastian Biedermann
Abstract:	As wireless networks continue to become more and more pervasive, the complexity of attacks targeting them increases accordingly. Although intrusion detection and prevention systems have become more robust, adversaries increasingly rely on covert channels to evade monitoring, particularly after an initial compromise has occurred. We introduce a covert channel that encodes hidden messages by modulating the transmission power of a device connected to a wireless local area network. We present a proof of concept (PoC) to establish such a covert channel. The performance of the prototype is evaluated under diverse experimental conditions: The evaluation covers different encoding schemes, different error correction strategies, different modulations of transmission power, and compares high-noise vs. low-noise environments. Our results demonstrate that the covert channel is feasible and resilient in different realistic wireless scenarios. Nevertheless, error correction strategies are mandatory and significantly prolong the transmission time. Although a viable distance of up to 18 meters could be measured, thick walls drastically reduce the possible distance. Based on empirical findings, we provide configuration recommendations and discuss implications for the detection and mitigation of this covert channel.
Download

Paper Nr:	103
Title:	Fine-Grained Access Control and Anonymization of Video Streams Using Attribute-Based Encryption
Authors:	Nouha Oualha and Christophe Janneteau
Abstract:	The proliferation of ubiquitous video surveillance and analytics systems, driven by sophisticated Artificial Intelligence (AI) object detection models, presents a critical challenge to individual privacy. Addressing the need for fine-grained, cryptographically enforced access control over sensitive visual data, this paper introduces an architecture and a functional demonstrator that employs Attribute-Based Encryption (ABE) to selectively obfuscate detected objects within video streams. The system operates by identifying objects of interest using an AI model and applying ABE to the corresponding regions of interest (RoIs) within the stream. Access to the obfuscated content is conditional, granted only to users whose cryptographic attributes satisfy a pre-defined access policy embedded in the ciphertext. The demonstrator facilitates the secure generation and distribution of ABE keys to various stakeholders. Furthermore, the accompanying client application validates the approach’s versatility by supporting on-demand encryption and decryption of diverse media types, including video, image, and PDF documents (treated as image data).
Download

Paper Nr:	107
Title:	Janus: A Novel Secret Key Generation Framework Combining W-State QKD and Post-Quantum Authentication
Authors:	Md. Rayhan Kabir Khan, Md. Fahim Shahoriar Titu, Istiak Ahammed, A. K. M. Iqtidar Newaz and Riasat Khan
Abstract:	In this work, we propose Janus, a hybrid architecture that integrates W-state-based QKD with CRYSTALS-Dilithium authentication to strengthen AES-256-GCM encryption. Unlike the GHZ state, the W-state offers improved robustness against qubit loss and decoherence, enabling reliable key generation in realistic quantum channels. Janus incorporates decoy states for eavesdropping detection and a privacy-preserving entanglement verification mechanism to ensure consistent key establishment. Dilithium signatures generate an authenticated pre-key, which is combined with the W-state-derived key and hashed to produce a 256-bit AES key. Our experimental evaluation on IBM’s ibm osaka backend demonstrates stable multipartite W-state generation, producing over 400 matching key bits from 512 transmitted entangled qubits.
Download

Paper Nr:	120
Title:	PRIVADA: Private User-Centric Data Aggregation
Authors:	Betül Aşkın Özdemir, Beyza Bozdemir, Ionut Groza and Melek Önen
Abstract:	In this work, we introduce PRIVADA, a maliciously secure data aggregation solution that uses MPC in the SPDZ framework. Unlike prior data aggregation schemes using MPC with/without SPDZ, PRIVADA supports multiple data customers while preventing inference of user participation and resisting collusions in real-world data aggregation applications. Moreover, our work guarantees user privacy and result privacy, in addition to input privacy. PRIVADA outperforms the state-of-the-art solutions by providing security against participating parties, including malicious data owners, aggregators, and data customers. Our proof-of-concept implementation also supports the new privacy-preserving data aggregation by combining malicious security, being available for multiple data customers, and ensuring strong privacy guarantees in large-scale deployments. The aggregation operation on the aggregator side becomes simpler with PRIVADA, and experimental results show a 12–15× speedup compared to the state-of-the-art. This confirms that malicious security and strong privacy guarantees can be achievable without sacrificing practicality.
Download

Paper Nr:	126
Title:	Efficient Second-Order Masked KECCAK Implementation without Online Fresh Randomness
Authors:	Juan Manuel Moreno-Cenizo, Eros Camacho-Ruiz, Macarena C. Martínez-Rodríguez and Piedad Brox
Abstract:	The KECCAK core is a fundamental cryptographic primitive widely used in post-quantum cryptography to ensure security against emerging quantum-computing attacks. Secure software implementations rely on masking schemes to mitigate side-channel leakage, but these protections introduce significant overhead, which is particularly impactful on resource-constrained microcontrollers and further exacerbated by micro-architectural leakages. To date, publicly available masked KECCAK implementations do not achieve both efficiency and robustness against such leakages. This work presents an optimized software implementation providing 1st-order side-channel protection, as well as the first efficiency-oriented 2nd-order implementation, surpassing previous solutions in performance. Both implementations target the ARM Cortex-M4 microcontroller and satisfy constraints to prevent micro-architectural leakages. A public repository provides the source code and a labeled dataset of 500K+500K traces, enabling reproducibility. Benchmarks show a 44.1% improvement over previously reported first-order implementations.
Download

Paper Nr:	152
Title:	Secure Multi-Hop QKD Protocol with Honest-but-Curious Relay Nodes
Authors:	Hiroki Yamamuro, Shusaku Uemura and Kazuhide Fukushima
Abstract:	Quantum key distribution (QKD) is a promising secret key exchange protocol that can replace the currently used public key cryptography, which is vulnerable to quantum computers. The main limitation of QKD is communication distance constrained by the attenuation in optical fibers. One solution to this limitation is classical relay using trusted relay nodes. This technology extends the communication distance by securely relaying a secret key from one node to the next using QKD-shared keys. The trusted relay node assumption incurs significant security costs. Therefore, relaxing this trust assumption is desirable. This paper proposes a QKD protocol that relaxes the reliability of relay nodes to honest-but-curious and enables long-distance communication through multi-hop transmission, which has no constraint on the number of relay nodes. Our protocol utilizes BB84 quantum states to transmit a bit sequence via relay nodes, and employs a secure classical channel implemented with post-quantum cryptography (PQC) to prevent relay nodes from deriving the shared secret key. We then analyze the security of our protocol against a possible attack involving the inference of the bit sequence by relay nodes, which arises from the relaxation of the reliability. We finally show that our protocol ensures the security against honest-but-curious relay nodes by discarding insecure secret keys with a high inference probability.
Download

Paper Nr:	161
Title:	The Equivalence of Two Quadratic Based IBEs
Authors:	George Teşeleanu
Abstract:	In this short note, we show that two identity-based encryption schemes, introduced by Joye and Zhao et al., which appear different, are essentially the same scheme. The only difference between them is that one prioritizes speed (Zhao et al.), while the other prioritizes bandwidth (Joye). We also show how to speed-up Joye’s scheme at the cost of adding at most one integer to the public key, thereby achieving a better encryption complexity, while having the same bandwidth requirements.
Download

Paper Nr:	184
Title:	Implementing Lightweight ARX Ciphers on a 6502-Based 8-Bit Platform: A Position Study on Performance and Code-Size Trade-Offs
Authors:	Dominik Madej and Filip Kruzel
Abstract:	In this article, we investigate whether modern lightweight block ciphers remain practical on a highly constrained 8-bit platform based on the MOS Technology 6502 processor. We use the Atari 65XE as the target system and implement two ARX ciphers, CHAM64/128 and SPECK64/128, in both C and hand-optimised assembly. The study compares execution time for key scheduling and single-block encryption, as well as executable size and RAM usage. Correctness is validated with known test vectors, while performance is measured using the platform’s native hardware timing facilities. The results show that manual assembly optimisation yields a close to one-order-of-magnitude speedup over compiler-generated C. Still, this improvement comes at the cost of a large increase in code size, mainly due to aggressive loop unrolling. Beyond reporting raw timings, the article positions 6502-class software cryptography as a useful stress test for understanding how algorithm structure, compiler limitations and low-level memory organisation interact in ultra-constrained environments.
Download

Paper Nr:	188
Title:	Proto-CBANet: A Statistical Metric Learning Framework for Imbalanced Intrusion Detection
Authors:	Chenxu Zhao
Abstract:	The detection of minority network attacks, such as User-to-Root (U2R), remains a challenge due to class imbalance. Traditional deep learning classifiers favor the majority class and exhibit large variations depending upon random initializations. We propose Proto-CBANet (Prototypical Class-Balanced Attention Network), which combines a 1D-CNN with an Average-Pooling Squeeze-and-Excitation block for feature recalibration, enabling distance-based reasoning in a latent metric space. To ensure statistical reliability, we evaluate our model across 60 random seeds. On NSL-KDD, Proto-CBANet achieves an average U2R recall of 61.07% ± 7.18%. To mitigate the trade-off between precision and recall for rare attacks, we further develop an advanced variant, HPN-GLF (Hyperspherical Prototypical Network with Global-Local Fusion). HPN-GLF integrates global learnable prototypes with hypersphere alignment, global-local semantic fusion, and dual-objective metric optimization. This advanced model reduces parameter count and FLOPs by 95% while significantly improving U2R precision from 12.95% to 59.76% and achieving the highest Macro F1 among compared methods. On NF-UNSW-NB15-v2, HPN-GLF attains a Macro F1 of 59.39% and high Shellcode recall (97.09%), indicating its promise for resource-constrained, imbalanced network security environments.
Download

Paper Nr:	191
Title:	Prose2Policy (P2P): A Practical LLM Pipeline for Translating Natural-Language Access Policies into Executable Rego
Authors:	Vatsal Gupta and Darshan Sreenivasamurthy
Abstract:	Prose2Policy (P2P) is an LLM-based system that translates natural-language access control policies (NLACPs) into executable Rego code, the policy language of Open Policy Agent (OPA). It provides a modular, end-to-end pipeline that performs policy detection, component extraction, schema validation, linting, compilation, and automated test generation and execution. The system is designed to bridge the gap between human-readable access requirements and machine-enforceable policy-as-code (PaC), with a focus on deployment reliability and auditability. We evaluate Prose2Policy on the ACRE dataset and demonstrate a 95.3% compile rate for accepted policies, with automated testing achieving an 82.2% positive-test pass rate and a 98.9% negative-test pass rate. These results indicate that the system produces syntactically robust and operationally testable policies suitable for Zero Trust and compliance-driven environments. However, LLM-generated authorization policies may introduce silent security risks, such as over-permissive access or missing constraints. Prose2Policy mitigates these risks through structured decomposition, schema-aware validation, and test-driven verification, improving the reliability of generated policies in security-critical contexts.
Download

Area 2 - Data Security And Privacy

Full Papers

Paper Nr:	31
Title:	SecGeo: Efficient Secure Geofencing
Authors:	Oliver Timothy, Jay Shinde and Hai-Van Dang
Abstract:	Geofencing is a widely adopted technology that enables virtual boundary enforcement for location-based applications. However, conventional geofencing methods pose significant privacy risks due to continuous location tracking. This paper presents SecGeo, a secure circular geofencing system leveraging homomorphic encryption to enable secure geofence containment checks while protecting user location privacy against not only attackers but also the other parties in the system. The proposed scheme integrates the Haversine formula with homomorphic encryption to compute distance securely without exposing users’ precise locations. Compared to the existing secure distance computation based on homomorphic encryption encryption in (ˇSeděnka and Gasti, 2014), our method reduces encryption overhead by reducing the number of encryption operations. Experimental evaluations demonstrate that the proposed SecGeo with Paillier homomorphic encryption reduces the system runtime by an average 46% and the communication cost between the device and the geofencing service by 46.7%. It also maintains lower latency (around 0.9 seconds per query vs 1.6 seconds per query when using the secure distance computation in (ˇSeděnka and Gasti, 2014)). The proposed scheme
Download

Paper Nr:	52
Title:	Authenticated Private Information Retrieval for Range Queries
Authors:	Hesham Youssef, Ying Cai and Soamar Homsi
Abstract:	We study the problem of enabling users to perform range queries over a cloud-managed database without revealing which records are retrieved, while also allowing users to verify the correctness of the returned results. Existing approaches support either query privacy through Private Information Retrieval (PIR) or query result authentication through Authenticated Data Structures (ADSs), but not both simultaneously for range queries. To address this problem, we present an Authenticated Private Information Retrieval (APIR) protocol for range queries with guarantees of soundness and completeness. We first propose a baseline scheme that combines PIR with Merkle Hash Trees (MH-trees), achieving authenticated query processing but incurring high communication and storage overhead due to redundant verification objects (VOs). To improve efficiency, we further introduce the APIR-tree, a novel authenticated data structure that embeds values at every node, enabling nodes to serve as VOs for their ancestors and eliminating redundancy. Simulation results show that the APIR-tree reduces storage overhead by up to 25× and communication costs by an order of magnitude while preserving strong privacy and authenticity guarantees.
Download

Paper Nr:	88
Title:	Locally Differentially Private Synthesis of Decentralised Heterogeneous Social Graphs via Spectral Embeddings and Bayesian Optimisation
Authors:	Manel Jerbi, Zaineb Chelly Dagdia and Sjouke Mauw
Abstract:	Synthesising heterogeneous social graphs in decentralised settings is challenging because clients observe only egocentric views and strict privacy rules prevent central access to raw links or content. We present a locally private spectral graph synthesis framework for federated social platforms and peer-to-peer applications. Clients compute Laplacian spectral embeddings and degree summaries of their ego graphs and perturb them locally using the High-Dimensional Spherical mechanism for continuous embeddings and the two-sided Geometric mechanism for integer-valued degrees. An honest but curious server receives only the perturbed summaries and reconstructs a synthetic graph through Bayesian optimisation of reconstruction parameters. All server-side processing operates solely on noisy data and therefore incurs no additional privacy cost. Content is modelled explicitly as nodes, preserving homophily and heterophily. We evaluate structural fidelity through degree heterogeneity, clustering, assortativity, and modularity, and assess downstream utility through link prediction and community detection. Experiments are repeated under multiple random seeds and results are aggregated for robustness. Across communication, citation, and social networks, the proposed method achieves competitive structural fidelity and downstream utility under user-level LDP, with particularly strong performance on assortativity and modularity at moderate to large privacy budgets. Computational trade-offs, scalability considerations, and potential threats are also discussed.
Download

Paper Nr:	102
Title:	A Privacy Model for Searchable Symmetric Encryption
Authors:	Manuela Horduna
Abstract:	Searchable symmetric encryption enables efficient keyword search over encrypted outsourced data, making it a key primitive for cloud storage. Since searchable symmetric encryption protocols inevitably leak side information through repeated searches, access patterns, update behaviour, and timing, an analysis of their privacy properties is crucial. Prior work along this direction considers leakage functions capturing what is revealed, and privacy guarantees often remain difficult to compare across schemes. We propose a privacy model for searchable symmetric encryption protocols that makes adversarial power a central parameter. We formalise adversaries capabilities and organise them into four classes based on two orthogonal dimensions, observational power (low vs. high) and privacy guarantee strength (minimal vs. strong). These classes induce four privacy levels giving rise to a privacy lattice, capturing implication relations between guarantees under increasingly powerful adversaries. This enables reasoning about how privacy guarantees change under different adversarial capabilities. We validate the privacy model by showing that representative symmetric encryption protocols schemes from the literature naturally instantiate the proposed privacy classes.
Download

Short Papers

Paper Nr:	41
Title:	FROG: Fragmentation for Obfuscated Geolocation
Authors:	Raul Rivero, Mohammad Kumail Kazmi, Ashwin Parameswaran, Premkumar Chandrasekar, Maurille Beheton and Soamar Homsi
Abstract:	Multimodal Large Language Models (MLLMs) have significantly advanced the retrieval of Publicly Available Information (PAI), enabling geolocation inference from visual content through natural language prompting. However, transmitting raw images to cloud-hosted APIs introduces critical privacy risks, particularly for sensitive actors who must conceal their identity or intent. We introduce FROG, a privacy-preserving framework that enables accurate geolocation predictions without exposing the original image to any external service. Instead of transmitting raw pixels, FROG extracts semantic visual features locally, distributes obfuscated partitions across multiple untrusted LLMs, and aggregates the resulting captions into a text-only geolocation query. No single provider ever receives a complete or coherent view of the source image or the user’s retrieval intent. Our evaluation across four commercial LLMs demonstrates that FROG achieves accurate city-level geolocation while maintaining strong privacy guarantees under a zero-trust threat model.
Download

Paper Nr:	36
Title:	A Performance Driven Decision Framework for Hybrid Homomorphic Encryption
Authors:	Hannah Meinhardt, Clemens Krüger and Dominik Schoop
Abstract:	To support privacy-preserving computation in resource-constrained environments, this paper presents a systematic performance comparison between homomorphic encryption (HE) and hybrid homomorphic encryption (HHE) on embedded edge hardware. While HE enables computations on encrypted data without decryption, its practical implementation remains challenging due to extensive computational and memory requirements. HHE alleviates this burden by having the client encrypt data with the lightweight symmetric cipher instead, while the data processor (server) handles the intensive transciphering into homomorphic format. We investigate the performance trade-offs between direct encryption with the homomorphic encryption scheme TFHE and a hybrid approach combining the Kreyvium stream cipher with TFHE. Testing on Raspberry Pi hardware shows HHE’s client-side advantages: Kreyvium’s pure encryption time is two times faster than TFHE, and it requires only 0.02% of its RAM. This efficiency comes at the cost of increased server-side processing during transciphering, shifting the computational burden from encryptors (clients) to data processors (servers). Based on measurements across different hardware configurations, we develop a decision framework for selecting HHE or HE based on data characteristics and operational requirements.
Download

Paper Nr:	153
Title:	A Privacy Preservation Framework for Vehicular Ad-Hoc Networks
Authors:	Eunice Naa Korkoi Hammond and Jan H. P. Eloff
Abstract:	The evolution of Vehicular Ad-hoc Networks (VANETs) and connected vehicles has advanced applications such as accident-avoidance alerts, navigation, and infotainment, enhancing intelligent transportation. However, critical cybersecurity challenges-particularly user privacy concerns-persist. These issues, ranging from location privacy and data exposure to GDPR compliance clashes, have hindered full-scale VANET deployment. Although international laws and Privacy Enhancing Technologies (PETs) have been adopted to mitigate these concerns, standardised frameworks and privacy-preserving mechanisms are still critically needed. To bridge this gap between regulation and implementation, this paper proposes a structured privacy-preserving governance and control framework designed to establish and satisfy privacy goals for VANETs. Termed the VANET Privacy Framework (VPF), this approach seeks to address these expressed privacy concerns.
Download

Area 3 - Distributed Systems And Architectures

Full Papers

Paper Nr:	73
Title:	Detection of Blockchain Address Poisoning Attacks in Rollups
Authors:	Kamil Kaczyński, Hideaki Miyaji and Aleksander Wiącek
Abstract:	An address poisoning attack is a special kind of phishing attack found on blockchains. It involves the use of lookalike addresses whose hexadecimal prefixes and suffixes closely resemble those of accounts with which the victim previously interacted. Prior literature has focused on detecting this attack mainly in the Ethereum network, which turned out to be very concerning as millions of dollars have been reported as stolen. This article presents a detection system for poisoning attacks in rollups, developed based on the characteristics of these attacks identified in prior research. Three Layer Two networks have been analysed: ZKsync, Arbitrum, and Optimism. More than 400k poisoning transfers have been found, with a total value of assets stolen reaching almost 200k USD, mostly in stablecoins. The results are not as striking as in Ethereum but the poisoning attack is indeed confirmed to be present in Arbitrum and Optimism. This work also provides a characteristic of the most successful example.
Download

Paper Nr:	79
Title:	SPARTON: Secure Dynamic Partition Scheme for Last-Level Cache
Authors:	Tejeshwar Bhagatsing Thorawade, Rishab Ravi, Varun Venkitaraman Iyer, Keerthisagar Kokkiligadda, Nirmal Kumar Boran and Virendra Singh
Abstract:	Modern multi-core computing systems with shared last-level caches (LLCs) are vulnerable to cache timing side-channel attacks. Techniques such as flush-based, contention-based, and occupancy-based attacks exploit the shared nature of LLCs to interfere with victim access patterns and potentially leak sensitive information. While static LLC partitioning can prevent such attacks, it often leads to poor cache utilization. Dynamic cache partitioning (DCP) schemes, such as utility-based cache partitioning (UCP), improve utilization by dynamically allocating cache ways based on utility. However, UCP introduces new side-channel vulnerabilities. Recent defense mechanisms like PASS-P, SCAM, and SCALE attempt to secure UCP but fail to address a new side-channel discovered in this work. Specifically, UCP’s transfer-on-a-miss policy enables attackers to infer how many new cache lines the victim accessed, revealing sensitive data. We propose SPARTON, a secure dynamic cache partitioning mechanism that eliminates this new side-channel. SPARTON transfers ways across all sets simultaneously, preventing attackers from observing set-level victim activity. To minimize performance overhead, SPARTON prioritizes transfer candidates that avoid back-invalidations and writebacks. Evaluated on 120 application pairs from the SPEC2006 and SPEC2017 benchmark suites, SPARTON secures UCP with an average performance overhead of just 0.6%. Furthermore, in 87.94% of transfers, SPARTON successfully selects candidates that do not trigger back-invalidations or writebacks, demonstrating the effectiveness of its candidate selection strategy.
Download

Paper Nr:	83
Title:	Confidential Key Management as a Service: Enhancing Availability and Isolation in Key Protection
Authors:	Ngoc Nhat Huyen Tran and Christof Fetzer
Abstract:	Key management is paramount to security concerns in cloud system. Key protection is vital to avoid data breaches due to the inadvertent or intentional disclosure of cryptographic keys. Hardware security modules (HSM) have been considered as a trustworthy entity to store keys and certificates. However, the systems using HSMs are difficult to achieve scalability because of the heavy workload of cryptographic operations, as well as the cost of the dedicated hardware. Based on the idea of key isolation by HSM, we introduce a confidential software-based HSM (ccHSM) designed to provide strong isolation and scalable key management without the need for dedicated hardware. ccHSM delivers cryptographic operations and secure secret storage to cloud-native services throughout lifecycle. To enable this, ccHSM adopts Intel SGX for a trusted execution environment and implementation of HSM features, PKCS#11-compliant key management. Furthermore, ccHSM is able to establish a secure, intra-host PKCS#11 communication with the target application to ensure low-latency and tamper-resistant interaction. Finally, we apply ccHSM for TLS key offloading, JSON Web Token authentication. Our evaluation shows that ccHSM doesn’t suffer substantial performance trade-offs while significantly reducing cost relative to conventional hardware and cloud-based solutions.
Download

Paper Nr:	211
Title:	Complying with the Right to Be Forgotten in Smart Mobility Data Sharing: A Delete-Only Redactable Consortium Blockchain
Authors:	Tin Tironsakkul, Pradip Kumar Sharma, Deepak Puthal, Vinayagam Mariappan and WonSik Hong
Abstract:	The immutability of blockchain, while ideal as an alternative solution for decentralised data sharing and enhancing data authenticity in smart mobility, clashes fundamentally with growing demands for data mutability driven by privacy regulations. This tension threatens the practical adoption of blockchain in privacy-sensitive smart mobility ecosystems. To reconcile this conflict, we propose a novel data redaction approach designed for consortium blockchains. Our solution introduces a fine-grained, controlled, and accountable delete-only redaction function. The proposed protocol enables data subjects or trusted authorities to initiate the deletion of sensitive message data within transactions using a digital signature. The redaction request is subsequently validated and approved by a subset of trusted authorities via an accountable threshold signature scheme, ensuring transparency and non-repudiation. Performance analysis of the proof-of-concept implementation demonstrates minimal overhead in efficiency and scalability, which illustrates that our proposed protocol satisfies requirements for smart mobility applications.
Download

Paper Nr:	237
Title:	Security-by-Design Reference Architecture for Data Governance in Healthcare Digital Twins
Authors:	Chiara Braghin, Stelvio Cimato, Andrea Marchesini, Fabio Palazzesi and Elvinia Riccobene
Abstract:	Digital Twins (DTs) are increasingly adopted in healthcare to enable real-time patient monitoring, predictive analytics, and decision support. However, their deployment relies on highly distributed and heterogeneous infrastructures, where sensitive medical data are continuously exchanged across multiple components. This raises critical challenges for secure data governance, including the protection of data across its lifecycle, compliance with regulatory frameworks, and interoperability among existing systems. Despite the growing body of research on healthcare DTs, existing approaches lack a systematic integration of security, privacy, and compliance mechanisms within a unified architectural model. In particular, no reference architecture currently addresses security concerns across data in transit, at rest, and in use in a comprehensive and reusable manner. In this paper, we address this gap by proposing a security-by-design reference architecture for data governance in healthcare Digital Twins. The architecture is derived through a combined research-driven and threat-driven methodology, explicitly linking requirements to data lifecycle threats and to relevant standards (e.g., GDPR, HIPAA, ISO). To assess feasibility, we implement a prototype based on a microservices architecture and evaluate it in a representative healthcare scenario. The results show how the proposed solution supports secure data management across data in transit, at rest, and in use, while preserving interoperability and compliance. It can thus be considered as a practical and reusable foundation for secure data governance in healthcare DTs.
Download

Short Papers

Paper Nr:	181
Title:	Privacy-Preserving Compliance on Public Ledgers via Selective Disclosure Authorization Schemes
Authors:	Supriya Khadka and Sanchari Das
Abstract:	Public distributed ledgers enforce integrity through radical transparency, creating tension with data minimization principles required for regulatory compliance. While Zero-Knowledge Proofs (ZKPs) offer a theoretical privacy solution, existing constructions often overlook adversarial constraints in smart contract environments. Specifically, the asynchronous decoupling of off-chain proof generation from on-chain submission introduces front-running and proof-reuse risks in public mempools. In this work, we formalize Selective Disclosure Authorization Schemes (SDAS), a cryptographic primitive for granular and revocable compliance checks on public ledgers without revealing the underlying witness. We define a security model for SDAS, introducing Ledger-Bound Attribute Unlinkability and Context-Aware Sender Binding to capture how valid proofs remain bound to their intended authorization context. To validate sender binding, we present ZK-Compliance, an Ethereum-based instantiation that operationalizes a user-controlled “Grant, Verify, Revoke” lifecycle. We implement the sender-binding component using a 14-constraint Circom circuit that anchors the zero-knowledge proof to the executing on-chain sender address. Our Sepolia evaluation confirms practical viability: browser-based proof generation executes in under 200 ms, and on-chain verification costs 240,512 gas, neutralizing proof reuse by different callers while preserving strict attribute privacy.
Download

Paper Nr:	186
Title:	Crowd-PoR: Proof-of-Reputation Consensus Protocol for Blockchain-Based Mobile Crowdsensing
Authors:	Zaina Maqour, Hanan El Bakkali, Driss Benhaddou and Houda Benbrahim
Abstract:	Mobile Crowdsensing (MCS) has emerged as a powerful paradigm for urban data collection, relying on the widespread availability of smart devices. However, ensuring the trustworthiness of users and collected data remains a critical challenge. While centralized reputation systems suffer from single points of failure, existing blockchain-based solutions rely on consensus mechanisms like PoW or PBFT, which are either too energy-intensive or scale poorly for resource-constrained mobile nodes. To address these limitations, this paper proposes Crowd-PoR, a novel Proof of Reputation consensus protocol specifically tailored for MCS. Instead of hash-power competition, Crowd-PoR utilizes reputation as the primary incentive and metric for block generation. We introduce a dynamic reputation evaluation model using our RBCrowd scheme to accurately reflect historical behavior. The block leader is selected randomly from among the top 20% of high-reputation participants, while consensus is achieved through a voting process conducted by the other 20% of nodes. Furthermore, we filter transaction per pair to effectively mitigate common malicious behaviors, such as bad-mouthing and Sybil attacks. Through security analysis and performance simulations, this study demonstrates that Crowd-PoR significantly reduces computational overhead, and maintains sub-minute consensus latency and high throughput for thousands of participants.
Download

Paper Nr:	53
Title:	Federated Attested Custody for Blockchain Assets under Post-Quantum Uncertainty
Authors:	Dominique Bolignano
Abstract:	The emergence of large-scale quantum computers introduces structural uncertainty regarding the long-term security of blockchain systems relying on elliptic-curve cryptography. During the transitional migration toward post-quantum schemes, public key exposure events create a selective attack surface: exploitation depends on a time-constrained race between quantum key recovery and blockchain confirmation dynamics rather than on instantaneous cryptographic collapse. This paper makes three contributions. First, we formally define public key exposure events and model the resulting quantum exposure surface as a stochastic process governed by on-chain settlement frequency. Second, we introduce a confirmation-aware exploitation model that captures the race between quantum recovery and network confirmation, and we derive a closed-form upper bound on expected extractable loss under bounded adversarial capacity. Third, we propose a federated, hardware-confined off-chain custody architecture that reduces exposure frequency and value concentration without modifying existing blockchain consensus rules, while relying on independently auditable and certifiable mediators to provide governable institutional trust. Together, these results transform transitional quantum risk from a binary collapse scenario into a parameterized and bounded engineering problem deployable within current blockchain infrastructures.
Download

Area 4 - Risks, Policies and Software Security

Full Papers

Paper Nr:	54
Title:	PCP-Mark: Software Watermarking with an Undecidable Problem
Authors:	David Schwarzbeck, Tobias Heineken and Michael Philippsen
Abstract:	Software watermarking embeds a message into programs to assert ownership. Existing techniques are prone to subtractive attacks that aim to remove the watermark. PCP-Mark achieves computability-theoretic full subtractive resilience. It embeds the watermark into the control flow of the program as instances of the Post Correspondence Problem (PCP). An attacker cannot remove a PCP-Mark without risking corruption of the program, even with full knowledge of the watermark location and method. (Only brute force attacks may succeed). Although a PCP-Mark is a dynamic watermark that unfolds in running loops, it resides in the static data of the program so that its presence can be proven without running the watermarked program. Heuristics keep the average runtime overhead of a PCP-Mark in benchmark programs below 4% despite of its theoretical complexity. PCP-Mark is available as an LLVM-pass that automatically adds the watermark to programs.
Download

Paper Nr:	129
Title:	MITRE ATT&CK-based Attack Chain Prediction Using Hybrid LSTM-Markov Models for Cyber Risk Assessment
Authors:	Mayank Raj, Nathaniel D. Bastian, Lance Fiondella and Gökhan Kul
Abstract:	Modern intrusion campaigns rarely unfold as isolated events; instead, adversaries chain tactics and techniques into multi-stage sequences that evolve over time. While existing detection systems can surface malicious activity after execution, they provide limited ability to anticipate an attacker’s next move. This work introduces a hybrid prediction framework that models adversary progression using MITRE ATT&CK technique transitions, combining long-range sequence learning from campaign-derived traces with short-range probability alignment from real-world intrusion telemetry. We construct a publicly releasable dataset of 4,849 ATT&CK-mapped campaign chains and 8,437 operational intrusion flows, and train a two-layer Long Short-Term Memory (LSTM) network to estimate technique-continuation likelihoods. A first-order Markov model provides empirical transition priors, and constrained beam search synthesizes forward attack paths from observed prefixes. A formally defined risk scoring model integrates per-technique likelihood (combining Exploit Prediction Scoring System (EPSS), Common Attack Pattern Enumeration and Classification (CAPEC), LSTM confidence, and Cybersecurity and Infrastructure Security Agency (CISA) Known Exploited Vulnerabilities (KEV) indicators), detectability (via D3FEND coverage), and OCTAVE-based impact into a continuous 0–10 scale. Across multiple training regimes, the model achieves 86% next-step prediction accuracy with Pearson r = 0.76 and Spearman ρ = 0.81 correlation against National Cyber Incident Scoring System (NCISS) reference severity under an 80/20 evaluation split, while maintaining 42.3% real-world transition coverage across 26,051 risk-ranked future chains. We confirm that all 4,849 chains from 33 campaigns share only N = 33 independent severity labels and report both chain-level consistency metrics and campaign-level error metrics accordingly. To support reproducibility, we release all datasets, code, models, transition matrices, and evaluation scripts. Results demonstrate that hybrid LSTM–Markov forecasting can predict plausible multi-step adversary evolutions and quantify their defensive urgency, shifting cyber defense from post-compromise reaction toward proactive risk-informed mitigation.
Download

Paper Nr:	193
Title:	Playbook Generation for Process Anomalies in Insider Threat Scenarios
Authors:	Kéren A. Saint-Hilaire, Nora Boulahia-Cuppens and Frédéric Cuppens
Abstract:	Insider threats represent a big challenge for organizations because insiders have legitimate access and knowledge of organizational processes. Effective defense requires both accurate detection of anomalous behavior and timely response actions. This paper proposes a unified approach that classifies log attributes by their importance for insider threat detection using process mining and an anomaly score to identify process deviations. To address the lack of automated responses, we propose to generate Incident Response (IR) playbooks for business processes using Named Entity Recognition (NER) on insider threat scenarios, combined with Part-of-Speech (POS) tagging and the RE&CT framework, a knowledge base of attack techniques and tactics inspired by MITRE ATT&CK that allows for categorizing IR techniques and actions, to map entities to response actions. The proposed approach is evaluated using coverage, redundancy, usefulness, and efficiency metrics, demonstrating its ability to produce comprehensive and compact response playbooks.
Download

Short Papers

Paper Nr:	68
Title:	Hybrid Vulnerability Detection: Combining Latent Semantic Analysis with LLM-Assisted Fuzzing via Context Distillation
Authors:	Kevin Nguyen and Dominik Schoop
Abstract:	As modern software systems grow in complexity, the number of vulnerabilities has grown rapidly, rendering traditional security strategies-including manual code reviews and rule-based static analyzers-increasingly resource-intensive and unable to keep pace with emerging threats. This paper presents a hybrid pipeline combining static vulnerability candidate selection using Latent Semantic Analysis with dynamic verification through Large Language Model-assisted fuzzing. The approach identifies semantically similar code patterns to known vulnerabilities without exact syntactic matches, and automates the generation of fuzzing harnesses and seed inputs to overcome coverage plateaus. We evaluate our pipeline on real-world C libraries (LibTIFF, libexpat, FreeType) with documented vulnerabilities. Results demonstrate that vulnerability candidates are ranked within the Top 3–8 % of all functions, and model-generated artifacts yield a 5-fold speedup over random fuzzing. The LLM Gemini 3 Pro achieves the highest harness generation success rate at 83 % (Pass@3). Our analysis reveals fundamental limitations: the mixed-vector problem dilutes vulnerability signals in multifunctional code, safety alignment mechanisms can inadvertently neutralize vulnerabilities through defensive code insertion, and static feature extraction without build context achieves only 19.93 % recall. The pipeline reduces manual effort while still requiring human validation.
Download

Paper Nr:	69
Title:	Static Code Analysis Output as a Proxy Indicator for Classifying JavaScript Obfuscation
Authors:	Alkın Alkan and Mehmet Tahir Sandıkkaya
Abstract:	This study investigates whether the output of Static Code Analysis (SCA) tools can serve as a proxy indicator for identifying obfuscated JavaScript. Two datasets are analyzed in their original and obfuscated forms: a controlled pilot dataset focused on common malicious patterns and a larger Kaggle dataset of approximately 1900 real-world JavaScript files. Four SCA tools are applied to both datasets: SonarQube, ESLint, NodeJsScan, and JSHint. The outputs are normalized into a unified feature schema, and pre/post-obfuscation deviance is quantified through issue counts and Kernel Density Estimation (KDE) of warnings normalized by file size and line count. Results show large and consistent increases in issue counts for ESLint and JSHint, substantial rises for SonarQube, and unstable shifts for NodeJsScan. The KDE analysis strengthens the proxy indicator’s effectiveness beyond raw counts by confirming that normalized warning intensity rises after obfuscation for SonarQube and JSHint. These patterns demonstrate that SCA warnings deviate under obfuscation and provide a practical proxy indicator for distinguishing obfuscated code from human-readable code.
Download

Paper Nr:	70
Title:	INFFLOW-RT+: Transforming Real-Time Risk into Adaptive Incentives
Authors:	Argiro Anagnostopoulou, Konstantinos Karlis and Dimitris Gritzalis
Abstract:	The growth of the Industrial Internet of Things (IIoT) introduces critical security challenges, particularly in managing dynamic network behavior. Traditional access control mechanisms evaluate permissions statically, failing to monitor indirect dependencies and data flows over time. To address this limitation, our prior work introduced INFFLOW-RT, a passive detection methodology that applies Bayesian inference and graph centrality to identify high-risk information flows in real time. However, passive detection cannot automatically mitigate active threats. This paper proposes INFFLOW-RT+, a closed-loop security architecture that translates real-time risk monitoring into automated enforcement actions. The proposed framework implements a fuzzy logic controller to map continuous risk metrics into proportional reputation adjustments, enabling the system to adapt to gradual behavioral shifts rather than relying on rigid binary rules. Furthermore, INFFLOW-RT+ integrates blockchain technology to execute these adjustments via smart contracts. This decentralized approach manages a token wallet for each network node, ensuring that the system enforces penalties and distributes incentives in an immutable and transparent manner. Finally, we present specific operational scenarios to illustrate the underlying logic of the framework and show how it dynamically adjusts node privileges in response to varying threat levels.
Download

Paper Nr:	100
Title:	CAPG-v2: Impact-Weighted, Multi-Path Vulnerability Prioritization over CVE-Centric Attack-Position Graphs
Authors:	Noufal Issa, Damas Gruska and Loubna Ali
Abstract:	Vulnerability management routinely relies on per-CVE (Common Vulnerabilities and Exposures) severity scores or exploitability scores, yet real intrusions are multi-step: attackers chain exploits across hosts to reach high-impact goals. Attack graphs capture these dependencies, but many prioritization methods still under-represent two practical realities: (i) overlap the same CVE can appear on many distinct attack paths, so patching it can block multiple routes; and (ii) impact heterogeneity different goals (e.g., domain admin vs. data exfiltration) imply different losses. Building on CAPG, a recent CVE representation designed to construct attack-position graphs, we introduce CAPG-v2: a lightweight extension that adds probabilistic semantics and goal impact annotations, enabling impact-weighted and overlap-aware prioritization. We formalize (a) a path-based score that aggregates across distinct goal-reaching paths and (b) an enumeration-free Monte Carlo marginal expected-loss reduction score that avoids explicit path enumeration. Rather than treating vulnerabilities as isolated items, CAPG-v2 supports patch prioritization as a graph-aware risk-reduction problem in which remediation decisions are guided by attacker behavior, shared attack routes, and the business impact of reachable goals. We further provide JSON artifacts and a reference evaluator to support replication and extension.
Download

Paper Nr:	109
Title:	HRes-Adapter: A Hierarchical Residual Parameter-Efficient Fine-Tuning Framework for Code Vulnerability Detection
Authors:	Alfredo Cuzzocrea, Md. Fahim Sultan and Mst. Shapna Akter
Abstract:	The proliferation of large-scale, pre-trained transformer models has significantly advanced the field of automated code vulnerability detection. However, the immense size of these models makes full fine-tuning for specific downstream tasks computationally irrational and resource-intensive. To address this challenge, Parameter-Efficient Fine-Tuning (PEFT) methods have emerged as a promising alternative, enabling adaptation with a minimal number of trainable parameters. This paper introduces HRes-Adapter, a comprehensive PEFT framework designed to enhance the performance of pre-trained models for code vulnerability detection while maintaining high parameter efficiency. Our approach integrates two key components module into the transformer architecture: a Hierarchical Residual (HRes) module that facilitates multi-scale feature refinement through a learnable information-mixing strategy and an Attention-Aligned Adapter (AAA) that applies context-specific refinements directly to the model’s attention calculation for enhancing downstream adaptation. We implement HRes-Adapter on the UniXcoder base model and evaluate its performance on a comprehensive suite of eight benchmark datasets for vulnerability detection. Our proposed HRes-Adapter achieves an AUC of 0.684 and F1 of 0.642 on Devign with only 0.958% trainable parameters, outperforming LoRA F1 of 0.621 in balanced Precision-Recall and surpassing TS-PEFT F1 0.658 in cross-dataset stability, while requiring 103× fewer updates than full fine-tuning. Across seven held-out benchmarks. These results establish the effectiveness of our proposed framework in achieving competitive performance compared to full fine-tuning and other state-of-the-art PEFT methods while offering substantial savings in computational resources.
Download

Paper Nr:	67
Title:	A Process-Integrated Multi-Criteria Framework for Information Security Investment
Authors:	Adam Filippidis and Constantinos S. Hilas
Abstract:	Information security investment decisions increasingly require structured evaluation under uncertainty and organizational constraints. Organizations frequently struggle to efficiently allocate limited resources while ensuring alignment between security scope and business objectives. This study proposes a process-integrated decision-support framework that combines Design Science Research with Multi-Attribute Utility Theory (MAUT) to address this challenge. The framework embeds structured scope elicitation within projectmanagement processes and links it to interval-based multi-criteria aggregation. It is evaluated through two analytically constructed scenarios representing low- and high-complexity environments. Results demonstrate that iterative scope construction combined with uncertainty-aware MAUT modelling enhances preference robustness, improves transparency of trade-offs, and reduces misalignment between investment alternatives and organizational requirements. The study contributes to the bridging of governance processes with formal decision analysis and by reframing information security investment as a dynamic scope-alignment problem rather than a static optimization task. Implications for research and practice are discussed.
Download

Paper Nr:	91
Title:	Semantic Feature Extraction from PE Headers for Malware Classification
Authors:	Fabiha Jalal, Sadia Tasnim Dhruba and Md Moniruzzaman
Abstract:	Portable Executable (PE) malware classifiers are routinely benchmarked on malware-only data or against small homogeneous benign corpora. Recent surveys (Ucci et al., 2019; Aboaoja et al., 2022) note that most existing PE-feature studies select attributes by availability or precedent rather than by security rationale, and large benchmarks such as EMBER (Anderson and Roth, 2018) group 2,381 features only by extraction source while Ahmadi et al. (Ahmadi et al., 2016) similarly use >1,800 features without semantic categorisation. We argue this practice misrepresents which signals a deployed detector actually relies on: the relative importance of the same 60 PE-header features changes substantially when benign samples are added to the evaluation, and shifts further as the benign corpus is diversified beyond a single source. To support this claim we organise 60 PE-header attributes into seven security-rationale categories (Structure Integrity, Execution Context, Memory Layout, Security Posture, Code Characteristics, Resource/Import, Anomaly Indicators) and evaluate on 1,263 MalwareBazaar samples plus 1,132 benign PE files (170 SysInternals + 962 DikeDataset (Iosif, 2021)). The 60-feature taxonomy matches EMBER’s F1 in a directly compared head-to-head while using 39× fewer features, a benign-source hold-out rules out a single-source artefact for the dominant feature, and Wilson-95% PPV bounds under deployment priors quantify what practitioners face at sub-1% malware prevalence.
Download

Paper Nr:	131
Title:	Managing Cybersecurity Compliance with Structured Guidance and Integrated Audit Support
Authors:	Mariana Andrade, João Rafael Almeida and José Luís Oliveira
Abstract:	As cybersecurity regulations such as ISO/IEC 27001 and the NIS2 Directive continue to expand in scope and complexity, organizations face growing challenges in translating regulatory obligations into actionable security policies and audit-ready evidence. Conventional compliance approaches rely on manual interpretation of regulatory texts, fragmented documentation repositories, and ad hoc audit preparation, introducing operational bottlenecks and exposing organizations to non-compliance risks. This paper presents a compliance management platform that operationalizes regulatory requirements through structured, expert-guided control implementation. It combines NLP extraction with human-supervised annotation to convert regulatory texts into machine-readable frameworks, enabling multi-framework management (ISO/IEC 27001:2022 and NIS2), control mapping, evidence tracking, and role-based audit workflows. In a task-based usability study with twelve participants, the platform scored 83.3 on the System Usability Scale (SUS), rated “excellent,” indicating that embedded guidance can reduce expertise barriers in cybersecurity compliance management.
Download

Area 5 - Security and AI/Machine Learning

Full Papers

Paper Nr:	46
Title:	TrusTEE: Storage-Centric Secure Federated Learning with Trusted Execution and Policy Enforcement
Authors:	George Popescu-Craiova and Maribel Fernandez
Abstract:	Federated Learning enables collaborative model training without sharing raw user data, yet remains vulnerable to privacy threats via shared gradients, model updates, and the hardware in which aggregation runs. Users must trust that centralised servers correctly aggregate updates and enforce data access policies. This paper introduces TrusTEE, an end-to-end hardware-backed confidential federated learning system that combines cloud Trusted Execution Environments (TEEs) with user-controlled, policy-enforced object storage. Unlike prior TEE-based federated learning work, TrusTEE treats cloud object storage as the primary communication substrate and trust boundary, using a minimal TEE-based aggregator and cloud-native role-based access control (RBAC) to separate roles. The system ensures that (1) raw client updates are never visible to the coordinator, (2) aggregation occurs entirely in hardware-encrypted memory, and (3) access to updates, aggregates, and global weights is enforced via strict identity-bound storage policies. We evaluate TrusTEE using the FLAIR dataset and show that confidential aggregation can be achieved while preserving standard federated learning dynamics.
Download

Paper Nr:	57
Title:	A Hierarchical Evaluation Framework for LLM-driven Threat Modelling Tools
Authors:	Josephine Marie Bakka, Andreas Kjeldgaard Brandhøj, Tobias Worm Bøgedal and Johannes Bjerva
Abstract:	AI adoption has accelerated with the rise of LLMs, and people within the field of security are increasingly exploring their practical value. Threat modelling is central to secure system development, yet it remains largely manual and the value of LLM-driven tools is unclear. Even when LLMs prove useful, selecting the right one can be more challenging than using it. This paper introduces a systematic evaluation framework for LLM-driven threat modelling tools to support tool selection, observing the general LLM-integration, governance risks, and allowing for comparison of tool output. Using Goal-Question-Metric, we derive evaluation metrics and show the value of the framework on a set of state-of-the-art LLM-driven threat modelling tools. The results show our metrics distinguish both performance and governance risks, providing a basis for organisations to ensure automation strengthens rather than burdens their threat modelling process.
Download

Paper Nr:	66
Title:	Generalizing across Networks: Evaluating Model Transferability for Intrusion Detection
Authors:	Miguel Silva, João Vitorino, Daniela Pinto, Ivone Amorim, Eva Maia and Isabel Praça
Abstract:	As cyber threats continue to evolve, attackers improve their techniques to optimize the effectiveness of their malicious activities. In response, there has been a significant increase in the implementation of Machine Learning (ML) and Deep Learning (DL) in Network Intrusion Detection Systems (NIDS). Choosing high-quality, representative datasets is a critical step in improving detection accuracy; however, many datasets have been available for a long time, some for over two decades, raising questions about the ability of ML and DL models to generalize across different and possibly outdated network scenarios. This study evaluates the generalization capability of models such as LGBM, RF, XGB, and LSTM, particularly in identifying previously unseen attacks. A significant challenge addressed is the inconsistency introduced by different flow exporters, which compute features differently, affecting model reliability. To mitigate this, HERA is used to rebuild datasets directly from original PCAP files. Additionally, we investigated the impact of feature selection on generalization and examined how performance changes when combining different datasets. When evaluated on alternative datasets, however, their behavior diverges, and in some cases, similar traffic patterns allow for reasonable generalization, while in others, the uniqueness of the dataset leads to significant performance drops.
Download

Paper Nr:	90
Title:	Assessing the Operational Impact of Poisoning Attacks over Augmented 3D Point Cloud Public Datasets for Connected and Autonomous Vehicles
Authors:	Marwan Lazrag, Badis Hammi, Lorena Gonzalez-Manzano and Joaquin Garcia-Alfaro
Abstract:	Poisoning attacks against public datasets lead to major concerns, such as (i) misclassification of perceived objects when the poisoned data is used for training and (ii) embedding of backdoors that may eventually be triggered later on, when specific conditions in the system apply over the learned models. Its impact over data augmentation models is unclear. While data augmentation reduces the likelihood of poisoning attack success, some valid questions remain. Is data augmentation affecting the impact of poisoning attacks? can it increase the number of poisoned samples or injected backdoors? We explore in this paper some of these questions. We assess the effects of augmenting poisoned 3D point cloud datasets and validate that poisoning is able to evade the sanitizing nature of augmentation techniques when using the concrete case of Generative Adversarial Network (GAN) techniques to exemplify the case of data augmentation processing. We also validate that poisoning propagates over the augmented datasets and perturbs the decision made by general-purpose classifiers, in the end. All the experimental material (including tools, datasets, and classifiers) is publicly available, to facilitate reproducibility and to foster further research in the topic.
Download

Paper Nr:	95
Title:	Dissecting Malicious VS Code Extensions: Characterization and Classification
Authors:	Kotaiba Alachkar, Dirk Gaastra, Karlo Zanki, Marc Ohm, Eduardo Barbaro and Yury Zhauniarovich
Abstract:	The Visual Studio Code (VS Code) extension ecosystem has increasingly been abused in software supply chain attacks. Prior work has studied extension security at scale and reported malicious cases, but has largely focused on identifying suspicious behavior rather than analyzing confirmed malicious extensions that were removed from the Marketplace, leaving them undercharacterized. In this paper, we analyze malicious VS Code extensions identified through Microsoft-maintained malicious and removed lists and construct a ground-truth dataset of 262 malicious extensions. We examine Marketplace metadata and extension activation behavior, analyze code obfuscation and entropy, assess VirusTotal (VT) results, infer malicious objectives through manual inspection, and perform a code similarity analysis to identify code reuse across extensions. Our analysis shows that malicious extensions typically execute early or unconditionally, frequently employ obfuscation, and are primarily used to establish persistent footholds for remote access and staged attacks rather than immediate destructive actions. We also find that 66.79% of confirmed malicious extensions are not flagged by any antivirus engine on VT, highlighting a significant detection gap. Based on the results of this analysis, we develop a baseline machine learning pipeline using 60 purely static features extracted from 262 malicious and 262 benign extensions that achieves an F1 score of 0.841 and an AUC of 0.919 in detecting malicious extensions, without relying on VT or Marketplace metadata.
Download

Paper Nr:	160
Title:	Bit-Flip Vulnerability of Shared KV-Cache Blocks in LLM Serving Systems
Authors:	Yuji Yamamoto and Satoshi Matsuura
Abstract:	Rowhammer on GPU DRAM has enabled adversarial bit flips in model weights; shared KV-cache blocks in LLM serving systems present an analogous but previously unexamined target. In vLLM’s Prefix Caching, these blocks exist as a single physical copy without integrity protection. Using software fault injection under ideal bit targeting, we characterize worst-case severity and identify three properties: (1) Silent divergence-13 of 16 BF16 bit positions produce coherent but altered outputs, indistinguishable from legitimate responses without a clean baseline. (2) Selective propagation-only requests sharing the targeted prefix are affected. (3) Persistent accumulation-no temporal decay occurs, so cumulative damage grows linearly with subsequent requests. Together, these constitute a threat profile distinct from weight corruption: silent divergence and selective propagation enable detection evasion; persistent accumulation then proceeds unchecked, yielding damage amplification bounded only by how long the block remains cached. A checksum-based countermeasure detects any single-bit corruption at scheduling time, bounding cumulative damage to one batch independent of the block’s cache lifetime, with negligible overhead. These results argue for integrity protection of prefix blocks before end-to-end exploitation is demonstrated.
Download

Paper Nr:	167
Title:	Adversarial Distance Metrics: A Threat to Fairness in Clustering-Based Decision Systems
Authors:	Shahzad Ahmad, Stefan Rass and Enes Sovtic
Abstract:	Clustering algorithms are increasingly deployed in high-stakes decision-making systems under the assumption that transparent, explainable methods using standard distance metrics are inherently trustworthy. We challenge this assumption by demonstrating a configuration-based attack exploiting ε-semimetric distance functions, which satisfy nearly all properties of a topological metric while allowing arbitrary manipulation of pairwise distances. Building on the mathematical framework of ε-semimetrics introduced in prior work, we make three novel contributions: (1) we formalize a threat model in which an insider adversary manipulates the distance function to induce discriminatory clustering outcomes against demographic groups, exposing a gap in current fairness auditing practices that focus on data integrity and algorithm transparency but overlook configuration integrity; (2) we provide empirical feasibility analysis demonstrating successful attacks on K-means, DBSCAN, and agglomerative clustering with 100% manipulation accuracy for sample sizes up to m = 130, with construction cost O(m6), and validate the attack on real census data (UCI Adult Income), demonstrating that Demographic Parity Difference increases from 0.200 to the theoretical maximum of 1.000 under attack; and (3) we propose concrete defenses including cryptographic metric commitment and statistical detection methods. The attack is most feasible for targeted discrimination affecting small groups (m < 50, under 30 seconds), precisely where aggregate fairness statistics lack power to detect violations. Our findings demonstrate the urgent need to expand the scope of the algorithmic audit to include distance-function verification.
Download

Paper Nr:	179
Title:	Fine-Tuning LLMs for Operational Phishing Email Detection
Authors:	Armand Florent Tsafack Piugie, Mathieu Valois, Emmanuel Giguet, Christophe Rosenberger and Philippe Chauvat
Abstract:	Phishing email attacks, a predominant and evolving cyber threat, necessitate detection systems that are not only accurate but also robust, efficient, trustworthy, and explainable. While Large Language Models (LLMs) offer transformative potential for semantic analysis in many areas, their practical deployment for phishing detection requires a holistic assessment beyond peak accuracy. This paper presents an operational phishing detection system that processes raw emails using a LLM and provides understandable explanations. We compared three fine-tuned LLMs among CamemBERT-base, XLM-RoBERTa, and Mistral 7B and we evaluate their performance, generalization capabilities, computational efficiency, and explainability on multiple significant datasets. Our results demonstrate that LLMs significantly outperform character-level models, with XLM-RoBERTa and Mistral 7B achieving over 99.4% accuracy. Critically, we reveal a fundamental trade-off: while Mistral 7B achieves marginally higher accuracy, its computational cost could be prohibitive for real-time deployment. In contrast, XLM-RoBERTa provides an optimal balance, offering state-of-the-art accuracy (99.46%), superior cross-dataset generalization (95.76% on an unseen corpus), and practical inference speed. Furthermore, by integrating SHAP-based explainability with a generative LLM, we validate that model decisions align with recognizable phishing indicators, bridging the gap between high performance and operational trust. This work provides a multi-dimensional framework for evaluating phishing detection systems and demonstrates that for most real-world applications, a multilingual, efficient encoder like XLM-RoBERTa represents the most viable and trustworthy choice.
Download

Short Papers

Paper Nr:	20
Title:	Packed and Unpacked Malware Detection by Means of Explainable Federated Machine Learning
Authors:	Giovanni Ciaramella, Fabio Martinelli, Antonella Santone and Francesco Mercaldo
Abstract:	The number of cybersecurity attacks has drastically increased in the last few decades. Although researchers and experts have proposed several methods to curb them, malicious actors continue to develop new techniques to perpetrate attacks. Among these techniques, attackers often use software known as a packer to obfuscate or compress executable code. Thus, it allows us to obtain a different file representation, making the analysis process more difficult. Given the widespread use of malware in companies and the need to avoid sharing sensitive data, this research article proposes a malware detection method based on Federated Machine Learning. In detail, we trained several models leveraging a publicly available dataset of Portable Executable and Object Linking and Embedding files from the Windows environment, which we converted into images using a Python script. To enhance the privacy and security of the model, we employed the zeroing norm aggregator, which zeroes to mitigate the risk of gradient leakage attacks. Moreover, we employed non-Independent and Identically Distributed data to represent a real-world scenario better. Once we concluded the training phase, we also evaluated the best model on original samples and on samples processed with three state-of-the-art packers (MPRESS, BEP, and GZexe) to assess its robustness. Specifically, the baseline accuracy of 0.881 increased slightly with MPRESS (+0.003) and GZexe (+0.004). However, the BEP-packed samples showed the greatest improvement, achieving an accuracy of 0.972. Finally, we applied the Gradient-weighted Class Activation Mapping++ algorithm to highlight the image regions that influenced the classification results.
Download

Paper Nr:	51
Title:	Generating Bijective 8-bit S-Boxes with Reinforcement Learning
Authors:	Lukas Surab and Pavol Zajac
Abstract:	Cryptographic S-boxes must balance security (low differential uniformity, high nonlinearity) with implementation cost. We formulate the S-box construction task as a Reinforcement Learning (RL) problem over reversible circuits and introduce a custom Gymnasium environment (SBoxEnv). The task is modelled as a Markov decision process (MDP), where the reward couples peak-focused improvements in differential uniformity (DU), nonlinearity (NL), and multiplicative complexity (MC) to flatten the differential and linear spectra of Walsh/Difference Distribution Table (DDT) distributions. The reward combines DU/NL improvements, spectral tie-breakers, stagnation control, circuit-length penalties, and a productivity-aware penalty for unproductive nonlinear gates. We evaluate three algorithms: PPO, A2C, and DQN across the gate sets {XOR, Toffoli}, {XOR, Fredkin}, and {XOR, Toffoli, Fredkin}. Our results are also compared with those obtained using similar techniques and the AES S-box. A diversity analysis over 100,000 generated candidates showed unique lookup tables and unique DDT/Linear Approximation Table (LAT) spectral fingerprints for all analysed S-boxes. RL-driven, non-template construction explores a much larger design space than algebraic recipes do, producing families of structurally diverse S-boxes and their implementations in the selected gate set.
Download

Paper Nr:	84
Title:	Explainable Multi-Objective Hybrid Metaheuristic Feature Selection for IoMT Intrusion Detection
Authors:	Selma Bououdina, Habib Louafi, Meriem Zerkouk, Neila Mezghani and Richard Hotte
Abstract:	The rapid growth of the Internet of Medical Things (IoMT) has significantly increased the volume and diversity of data generated by connected medical devices, raising important security challenges for intrusion detection systems (IDS). However, the high dimensionality of network traffic data can negatively affect IDS performance and increase computational cost. Feature selection plays a critical role in addressing this issue by identifying the most relevant features for accurate detection. Nevertheless, exploring all possible feature subsets is computationally intractable due to the exponential search space. In this paper, we propose a hybrid metaheuristic-based feature selection approach for IDS in IoMT environments. The proposed method formulates feature selection as a multi-objective optimization problem that simultaneously considers classification performance and subset compactness. Several machine learning algorithms are used to evaluate the quality of the selected feature subsets. Experimental results demonstrate that the proposed approach is capable of selecting informative and compact feature subsets while maintaining competitive detection performance and reducing computational cost. Furthermore, an explainable artificial intelligence (XAI) analysis based on SHAP is employed to interpret the contribution of the selected features to the classification decisions, improving the transparency and interpretability of the model.
Download

Paper Nr:	110
Title:	LogSanitizer: Defending LLM-Integrated SOCs against Backdoor Triggers Delivered through Firewall Logs
Authors:	Leszek Wronski and Bogdan Ksiezopolski
Abstract:	The integration of Large Language Models (LLMs) into Security Operations Centers (SOC) introduces a novel cross-layer attack surface that has not been previously studied: adversary-generated log injection (AGLI), where an external attacker sends crafted network packets that are blocked and logged by the firewall, producing structured log entries with attacker-controlled fields. When these logs are forwarded by the SIEM to a trojaned LLM for analysis, the model recognizes the embedded trigger and generates malicious recommendations disguised within legitimate security advice. We propose LogSanitizer, a family of input sanitization defenses operating at two levels: a pre-prompt log-transformation pipeline that disrupts trigger patterns in the structured log representation, and a post-tokenizer perturbation strategy that corrupts trigger-bearing token configurations before they reach the model. We evaluate both approaches against a multidimensional backdoor embedded in Foundation-Sec-8B. Through iterative refinement we addressed three adversarial phenomena discovered during defense development: Out-of-Distribution (OOD) failures from type-altering transformations, pattern collision via entropy loss, and the truncation trap in temporal jittering. The final pre-prompt pipeline achieves complete trigger neutralization (0.0% attack success) with 100.0% task utility retention, while post-tokenizer token substitution at a 10% perturbation rate achieves comparable results on a smaller evaluation sample.
Download

Paper Nr:	123
Title:	C-STAR: Cost-Aware Adaptive Learning under Concept Drift for Android Malware Detection
Authors:	Nahee Kwon, Kyoungmin Roh, Youngsup Hwang, Seong-je Cho and Boojoong Kang
Abstract:	Android malware detection models degrade over time as evolving applications and attack behaviors induce concept drift. Existing mitigation strategies typically rely on periodic or cumulative retraining, but exhaustive updates incur substantial cost and may destabilize previously learned knowledge. We reformulate drift-aware model maintenance as an operational decision problem that determines when to update and how much data to use under resource constraints. Based on this view, we propose C-STAR (Cost-Aware Selectively Triggered Adaptive and Robust Learning), which combines structural drift monitoring, informative sample prioritization, and cost-aware update optimization. In contrast to prior approaches, C-STAR explicitly incorporates forgetting risk into the update objective. Experiments on the AndroZoo dataset show that C-STAR achieves performance comparable to cumulative retraining while substantially reducing update data usage, computational overhead, and forgetting risk. Our results indicate that reliable malware detection under drift can be maintained through selective cost-aware adaptation rather than exhaustive retraining.
Download

Paper Nr:	136
Title:	Federated Learning for Malware Image Classification under Data Heterogeneity
Authors:	Victor Taiwo, Cemal Nişan, Muhammad Athallah, M. Emre Gürsoy and Öznur Özkasap
Abstract:	Malware image classification (MIC), in which binary files are converted into visual representations and deep neural networks are trained to identify malware families, has recently emerged as an effective approach for malware detection. Federated learning (FL) enables malware image classifiers to be trained collaboratively by multiple clients without sharing raw data. However, a fundamental challenge in FL is data heterogeneity (non-IID), which is particularly relevant but understudied in the malware context. In this paper, we address this gap via three contributions. First, we present an empirical study of federated MIC under varying degrees of data heterogeneity across three popular malware image datasets (MalImg, Virus-MNIST, and MalNet). We show that while FL achieves strong accuracy under IID conditions, both accuracy and training stability degrade as data becomes increasingly non-IID. Second, we introduce a set of round-level statistics that characterize the data contributed by participating clients in each FL round, and demonstrate that class entropy (diversity of class labels present in a given round) exhibits strong positive correlation with global model accuracy. Third, leveraging this insight, we propose an entropy-aware client selection strategy that filters candidate client subsets based on class entropy, and show that our strategy outperforms standard random selection. Our results highlight the role of client composition in federated MIC and underscore the importance of distribution-aware classifier training under data heterogeneity.
Download

Paper Nr:	137
Title:	Preserving Membership Privacy via Risk Score Guided Knowledge Distillation
Authors:	Radia Kassa, Kamel Adi and Abdelkamel Tari
Abstract:	The increasing deployment of deep learning models has raised important concerns regarding data privacy. In particular, Membership Inference Attacks (MIAs) aim to determine whether a specific data sample was used to train a model, potentially exposing sensitive information. To address this issue, we propose Risk Score guided Knowledge Distillation (RS-KD), a defense mechanism that dynamically adapts the distillation process according to a privacy risk score estimated from the output characteristics of a teacher model. This risk estimation enables the identification of samples that are more vulnerable to MIAs, allowing the framework to selectively regulate the uncertainty introduced during distillation. Experimental results on benchmark datasets demonstrate that RS-KD significantly reduces the effectiveness of black-box MIAs while preserving predictive performance, achieving a favorable privacy–utility trade-off.
Download

Paper Nr:	143
Title:	Size Does Matter: The Impact of Embedding Models and Sizes on Spam Email Classification
Authors:	Malte Josten, Gérald Kämmerer, Arne Kummerow and Torben Weis
Abstract:	Spam and phishing emails remain a major cybersecurity challenge, even after decades of research into reliable detection methods. Modern ML-based spam filters typically rely on text embeddings to represent email content, yet the choice of embedding model and size is often treated as secondary. This work empirically compares a diverse set of sentence embedders to assess how model type and embedding dimensionality influence downstream email spam classification. Using both classical and ML-based classifiers, we evaluate performance across multiple embedding configurations. Our results show that embedder choice - especially embedding size - substantially affects classification performance and generalisation. We observe performance differences of up to 13% overall, alongside variations of 25% in misclassified spam and 10% in misclassified ham across embedders. These findings highlight that embedding models are not interchangeable; rather, their deliberate selection is just as critical as choosing the right classifier when designing AI-based spam detection pipelines.
Download

Paper Nr:	197
Title:	Graph2TTP: Knowledge Graph-Guided Paragraph-Level TTPs Identification from Cyber Threat Intelligence Reports
Authors:	Patrick Zounon, Yufei Han, Michel Hurfin and Frédéric Majorczyk
Abstract:	Mapping unstructured Cyber Threat Intelligence (CTI) reports to the MITRE ATT&CK framework is critical for proactive defence but remains a manual, time-consuming process. Existing automated approaches either rely on black-box language models that lack interpretable reasoning or brittle, rule-based knowledge graphs that fail to scale. To bridge this gap, we propose Graph2TTP, a novel neural-symbolic framework for automated, paragraph-level Tactic, Technique and Procedure (TTP) identification. Graph2TTP leverages the zero-shot comprehension of Large Language Models (LLMs) to automatically extract entities and relations from extensive CTI narratives, constructing rich, localized Cyber Security Knowledge Graphs (CSKGs). We encode these symbolic structures into dense embeddings and process them via a multi-label Edge-featured Graph Attention Network (EGAT). This hybrid architecture delivers both the high-fidelity accuracy of neural networks and the transparent, verifiable reasoning paths required by security analysts. To facilitate rigorous evaluation, we curate and release APTCTI, a real-world dataset comprising 690 Advanced Persistent Threat reports spanning over 18,087 paragraphs. Extensive evaluations across multiple datasets demonstrate that Graph2TTP outperforms state-of-the-art neural baselines (e.g., CySecBERT) by approximately 20% in F1 score, establishing a robust new standard for accurate and interpretable threat intelligence analysis.
Download

Paper Nr:	199
Title:	Robustness Evaluation of Graph Neural Network-Based Network Intrusion Detection Systems against Adversarial Flow Injection
Authors:	Matthieu Mouzaoui, Yufei Han, Gregory Blanc, Gabriel Rilling and Michel Hurfin
Abstract:	Graph Neural Networks (GNNs) enhance Network Intrusion Detection Systems (NIDS) by leveraging structural relationships in network traffic, but this reliance introduces new vulnerabilities. In this paper, we propose a query-based black-box flow injection attack against GNN-based NIDS, where adversaries inject crafted traffic flows into existing communications to evade detection. Our experiments demonstrate that these constrained attacks are highly effective, frequently succeeding with a minimal budget of injected flows. Analysis reveals that local graph structural properties primarily determine attack success. Additionally, the attack demonstrates strong transferability across various GNN architectures, confirming the viability of a black-box threat. These findings highlight critical vulnerabilities in GNN-based NIDS, even when adversaries are restricted to purely additive flow injections.
Download

Paper Nr:	45
Title:	ORCHID-FL: On-Device Relationship-Aware Contextual Handling of Information via Decentralized Federated Learning
Authors:	George Popescu-Craiova
Abstract:	Mobile devices store highly sensitive personal data, yet privacy failures often arise not from technical compromise but from mismatches between user intent and sharing interfaces that offer little contextual nuance. We propose ORCHID-FL, an on-device, language-mediated privacy agent that combines tag-grounded perception, a promptable LLM that proposes sharing decisions, and a deterministic rule engine that retains final authority. To allow the agent to adapt to evolving social and institutional norms without centralising sensitive data, ORCHID-FL is designed to participate in a federated fine-tuning loop using parameter-efficient (LoRAstyle) updates. We position this paper as an architectural proposal: we present the system design, a hybrid LLM + deterministic-rule decision pipeline, and a threat model that treats the LLM and the federation channel as untrusted. We provide preliminary empirical evidence on the perception layer by comparing CNN-tag, multimodal-tag, and end-to-end multimodal architectures on 200 VizWiz-Priv images annotated by the author, and find that structured tag-grounded pipelines achieve the highest decision agreement and the lowest unintended disclosure rate. Federated fine-tuning evaluation, multi-annotator validation, and quantification of the differential-privacy/utility trade-off are deferred to future work.
Download

Paper Nr:	62
Title:	Large Language Model Vulnerabilities
Authors:	Meda Račaitytė, Hélder Bastos, Ricardo Ribeiro and Luís Nunes
Abstract:	Large language models are increasingly being deployed in safety-critical domains, yet remain vulnerable to jailbreak attacks that circumvent safety alignments. This systematic review synthesizes empirical jailbreak research published between 2024 and 2025, using a PRISMA-guided search protocol, followed by BERTopic-based topic modeling. The analysis identifies eight main jailbreak categories: optimization-based, genetic/evolutionary, iterative refinement, semantic/persuasion-based, decomposition, context/generation-level, visual/encoding and fuzzing attacks, and characterizes their effectiveness, efficiency, and transferability across open-source and proprietary models, including Llama-2/3, Vicuna, GPT-3.5/4, Claude, Gemini, and DeepSeek-V3. Results show that simple configuration and context-level attacks can match the near-perfect attack success rates of sophisticated white-box optimization methods on models such as Llama-2, while requiring far fewer queries and no parameter access, highlighting a gap between research focus and practical threat severity. The review further identifies five recurring vulnerability mechanisms: representation-level gaps, execution-priority manipulation, semantic fragmentation, gradient-space exploitation and persuasion susceptibility, and documents family-specific vulnerability patterns, with open-source Llama-based models consistently more exposed than safety-enhanced architectures such as Claude. Diverse methods, uneven focus on models and publication bias limit how broadly results apply. Nonetheless, the review reveals that weaknesses in safety alignment persist across successive LLM generations, urging that effective defenses must address all eight attack categories rather than isolated techniques.
Download

Paper Nr:	98
Title:	Logfusion: A Lightweight Preprocessing Pipeline for Anomaly Detection
Authors:	S. M. Aminur Rahman, Abu Barkat Ullah, Masoud Mohammadian and Yibeltal F. Alem
Abstract:	The increasing frequency and sophistication of cyber-attacks continue to expose vulnerabilities in modern digital infrastructures. System logs provide a rich source of evidence for detecting such threats, but raw logs are noisy, heterogeneous, and imbalanced, making them inconvenient for direct machine learning applications. This paper presents LogFusion, a lightweight preprocessing pipeline that fuses contextual embeddings of log messages with engineered behavioural indicators (e.g., temporal windows, event rarity, and log-level dynamics) to produce machine learning-ready features for unsupervised anomaly detection algorithms. Evaluated on three public benchmark datasets (HDFS, BGL, and OpenStack), LogFusion increases anomaly detection accuracy from 51–61% (raw Bag-of-Words baseline) to 84–98% across all detectors, with consistently high recall (>0.95 on HDFS and OpenStack). On HDFS, LogFusion-OCSVM achieves 87% F1-score, outperforming LogBERT (82.32%) and other deep learning baselines. Runtime profiling confirms end-to-end processing under 13 seconds on CPU-only hardware, demonstrating practical feasibility for resource-constrained deployments in SMEs and IoT/edge environments.
Download

Paper Nr:	121
Title:	Short-Term Temporal Behavioral Drift in Smartwatch User Authentication: A Case Study Using Apple Watch Sensor Logs
Authors:	Maharage Nisansala Sevwandi Perera, Takeshi Kawamoto, Allam Shehata, Franziska Zimmer, Ryosuke Kobayashi, Mhd Irvan, Rie Shigetomi Yamaguchi and Yasushi Yagi
Abstract:	Currently, user authentication using smartwatch sensor logs has gained attention for its ability to detect user-specific behaviors and its potential for continuous, unobstructed authentication. However, a key challenge in behavioral authentication is the impact of temporal changes, which can degrade accuracy. Such changes can result from environmental factors, emotional states, fatigue, or the passing of time. This study investigates short-term temporal behavioral drift and its impact on user authentication performance using Apple Watch sensor logs from 100 users. We collected approximately three hours of behavioral data per participant and analyzed the changes in behavior at 10-minute intervals. We evaluated four widely used classifiers, Random Forest, XGBoost, Logistic Regression, and SVM, to compare their robustness against short-term behavioral changes. The experimental results show noticeable performance variation between temporal windows, highlighting the presence and degree of behavioral drift. Despite temporal changes, our models based on Random Forest and XGBoost consistently achieved strong accuracy in distinguishing users and identifying genuine users with 80% probability and adversarial access with 80% probability. The findings provide insight into model reliability over short durations and offer practical guidance for designing robust smartwatch-based behavioral authentication systems.
Download

Paper Nr:	134
Title:	Fighting Cybercrime: Identifying Ransomware Families Using API Sequences and Retrieval-Augmented Ransom Note Analysis
Authors:	Gheorghe Balan, Ciprian-Alin Simion and Dragoş Teodor Gavriluţ
Abstract:	Ransomware attribution is increasingly relevant for cybercrime investigation and incident analysis. This paper studies ransomware family identification based on API-call sequences and campaign linkage using binary decision trees, random forests, and XGBoost. The proposed API-sequence based approach achieves an attribution accuracy of approximately 99%, showing that such models capture both family-specific behavior and cross-family similarities. To complement this behavioral perspective, we introduce a retrieval-augmented generation (RAG) workflow over a corpus of historical ransom notes. This component provides an explainable textual signal derived from extortion language, contact channels, negotiation templates, and family-specific wording. Historical notes are embedded and indexed with family metadata, while new notes are analyzed through semantic retrieval and structured large language model reasoning. Evaluated qualitatively, the RAG layer supports grounded analyst-facing attribution by confirming family labels, surfacing plausible aliases or rebrands, and identifying cases in which template reuse makes attribution ambiguous.
Download

Paper Nr:	140
Title:	Enhanced Cyber Threat Severity Prediction Leveraging CVSS and BERT-Based Language Models
Authors:	Mohamed El Amine Bekhouche and Kamel Adi
Abstract:	Accurate cyber threat severity prediction is essential for proactive risk management and efficient resource allocation. Traditional approaches mainly rely on the Common Vulnerability Scoring System (CVSS), which inadequately handles threats exploiting multiple vulnerabilities simultaneously. In this paper, we propose a methodology for cyber threat severity prediction. First, we introduce two aggregation strategies, sequential and parallel exploitation, to compute overall threat severity scores from CVSS metrics while considering multiple exploited vulnerabilities. Second, we propose a predictive model based on Large Language Model (LLM) encoders, namely BERT and the cybersecurity-oriented SecureBERT+. The model processes vulnerability descriptions to generate contextual embeddings, applies attention mechanisms to highlight informative features, and uses Long Short-Term Memory (LSTM) layers to capture contextual dependencies before predicting the threat severity score. Experimental results show improved predictive accuracy and lower error rates, particularly with SecureBERT+.
Download

Paper Nr:	180
Title:	MOSAIC-FL, a Micro-Service Based Privacy-Preserving Framework with Application to Genomics
Authors:	Paul Largillier, Karl Paygambar, Cédric Gouy-Pailler, Vincent Meyer, Mallek Mziou and Oana Stan
Abstract:	Security and privacy are primordial requirements for Federated Learning (FL) especially in fields such as healthcare and genomics when sensitive information has to be analyzed. Our FL framework is designed to address these challenges while proposing a modular, flexible and micro-service architecture. More precisely, it integrates an efficient gRPC communication layer and a Finite State Machine to ensure robust component synchronization and threat detection while based on a fault-tolerant secure aggregation protocol using a Threshold variant of CKKS homomorphic cryptosystem. This allows blind model aggregation by an orchestration server, requiring a minimum of t-out-of-N active clients for decryption while minimizing communication overhead thanks to both cryptographic and network protocols. We demonstrate the framework’s effectiveness through diverse use cases, ranging from standard image recognition (EMNIST) to complex genomic classification including breast cancer subtyping on TCGA, evaluating system performance across different threshold value and model scales.
Download

Paper Nr:	210
Title:	Detecting Concept Drift in Evolving Malware Families Using Rule-Based Classifier Representations
Authors:	Tomáš Kalný, Martin Jureček and Mark Stamp
Abstract:	This work proposes a structural approach to concept drift detection in malware classification using decision tree rulesets. Classifiers are trained across temporal windows on the EMBER2024 dataset, and drift is quantified by comparing extracted rule representations using feature importance, prediction agreement, activation stability, and coverage metrics. These metrics are correlated with both accuracy degradation and data distribution shift as complementary drift indicators. The approach is evaluated across six malware families using fixed-interval and clustering-based windowing in family-vs-benign and family-vs-family settings, and compared against RIPPER and Transcendent baselines. Results show that fixed two-month windowing with feature-level Pearson correlation is the most reliable configuration, being the only one where all family pairs produce positive drift–accuracy correlations. The methods are complementary - no single approach dominates across all pairs.
Download

Paper Nr:	213
Title:	Zero-Query Black-Box Adversarial Attacks Using Transferability on Object Detection Models
Authors:	Kohei Kawasumi and Kazuki Yoneyama
Abstract:	Non-maximum suppression (NMS) is commonly used to filter redundant detection results in object detection models. Several white-box adversarial attacks to induce false positives by circumventing the NMS and increasing the number of false bounding boxes have been studied. However, since white-box attacks require internal parameters of the target model, actual viable environments are limited. This paper aims at an adversarial attack in a more feasible black-box environment threat model without internal parameters of the target model. Specifically, we propose a zero-query attack that exploits transferability such that we generate a universal adversarial perturbation (UAP) by the ensemble of multiple surrogate models that are white-box accessible and use the UAP to attack the target model in the black-box environment. We show that by increasing the number of surrogate models, we can achieve the false positive rate of the target model up to 84.4% in COCO2017 dataset and 82.9% in BDD100K dataset for targeting YOLOv8, and 87.2% in COCO2017 and 79.7% in BDD100K for targeting YOLOv9.
Download

Paper Nr:	232
Title:	Adversarial Co-Evolution of Malware and Detection Models: A Bilevel Optimization Perspective
Authors:	Olha Jurečková, Martin Jureček, Matouš Kozák and Róbert Lórencz
Abstract:	Machine learning-based malware detectors are increasingly vulnerable to adversarial examples. Traditional defenses, such as one-shot adversarial training, often fail against adaptive attackers who use reinforcement learning to bypass detection. This paper proposes a robust defense framework based on bilevel optimization, explicitly modeling the strategic interaction between a defender and an attacker as an adversarial co-evolutionary process. We evaluate our approach using the MAB-malware framework against three distinct malware families: Mokes, Strab, and DCRat. Our experimental results demonstrate that while standard classifiers and basic adversarial retraining often remain vulnerable, showing evasion rates as high as 90%, the proposed bilevel optimization approach consistently achieves near-total immunity, reducing evasion rates to 0–1.89%. Furthermore, the iterative framework significantly increases the attacker’s query complexity, raising the average cost of successful evasion by up to two orders of magnitude. These findings suggest that modeling the iterative cycle of attack and defense through bilevel optimization is essential for developing resilient malware detection systems capable of withstanding evolving adversarial threats.
Download

Paper Nr:	234
Title:	Towards AI-Guided Security Hardening of Industrial Systems Based on IEC 62443
Authors:	Christophe Ponsard and Jean-François Daune
Abstract:	Industrial systems are increasingly exposed to cyber threats, requiring stronger design methodologies to ensure resilience and security. In critical infrastructures, compliance with standards such as IEC 62443 is essential for ensuring security throughout the system lifecycle. This paper investigates a secure-by-design engineering process focusing on the initial architectural design and examines the role that AI-powered agents can play in supporting it, as well as the conditions required for their effective and reliable use. This study experiments with an open-source agentic framework to generate zone-and-conduit architectures from an initial system blueprint. A multi-agent workflow is implemented, including agents responsible for architecture generation, compliance verification against IEC 62443 requirements, and security stress testing using attack tree analysis. A drinking water utility case study, subject to NIS regulatory constraints, is used to compare AI-assisted and manually produced architectures. The results highlight both the benefits and limitations of AI-assisted design, leading to the identification of recommendations and open issues for further research, with potential applicability beyond the industrial domain.
Download

Area 6 - Security And Privacy in Emerging Scenarios

Full Papers

Paper Nr:	55
Title:	Birds of a Feather Flock together: Network-Based Detection of Coordinated Disinformation Campaigns on Telegram
Authors:	Panteleimon Tsagkarakis, Emmanouil Papadogiannakis and Evangelos Markatos
Abstract:	In recent years, disinformation has increasingly proliferated across social networks. The firing of fact-checkers from Meta and the disbanding of Twitter’s Trust and Safety Council suggest that this trend will continue to escalate. While disinformation sources (e.g., social network accounts) can sometimes be identified, new accounts emerge daily, making tracking a moving target. In this work, we propose a graph-based methodology to discover previously unknown Telegram accounts that spread disinformation. Starting from a set of verified disinformation groups, we examine their interconnections and identify new accounts that contribute to the dissemination of false narratives. Our approach is language-agnostic, as it relies solely on structural relationships between accounts rather than analyzing their message content. Using this approach, we identify 37 previously unknown disinformation channels (a threefold increase). We demonstrate that Telegram channels display extremely dogmatic behavior with up to 80% of messages being labeled as propaganda. Our findings reveal that misinformation on Telegram spreads within tightly interconnected clusters, in some cases, with over 86K identical messages being shared to multiple channels, suggesting coordinated disinformation campaigns.
Download

Paper Nr:	60
Title:	Oblivious Complex Queries on Variable-Length Strings
Authors:	Mariagiovanna Rotundo, Giuseppe Persiano and Paolo Ferragina
Abstract:	In this paper, we study the problem of storing and searching in datasets of variable-length strings, a core primitive in key-value stores, (graph) DBs, and search engines. However, enabling such search capabilities in ORAM scenarios, where data are stored on an honest-but-curious server, remains challenging. We address this problem by proposing a practical design that combines Ring ORAM (Ren et al., 2015) to hide access patterns to outsourced data, with a Patricia trie (Ferragina and Grossi, 1999; Ferragina et al., 2025) for space-efficient search over variable-length strings. The resulting scheme supports search over variable-length string datasets in an ORAM scenario, while retaining efficient storage and access both on the client and the server. We evaluated our scheme on datasets having size up to 273 GB, showing that it supports complex string queries, with only 2 Ring ORAM accesses on the server, incurring a client-server communication cost below 3 MiB, a client memory footprint of at most 200 MB, and negligible client computation time per query. Although we assume bounded-length strings, the bound is high enough to handle most practical use cases.
Download

Paper Nr:	87
Title:	Achieving Obliviousness in Compressed Data Structures
Authors:	Mariagiovanna Rotundo and Paolo Ferragina
Abstract:	Oblivious RAM (ORAM) is a general-purpose storage technique that hides memory access patterns, protecting data, queries, and results from untrusted honest-but-curious servers. While traditional ORAM supports only basic reads and writes, recent works have addressed the problem of enabling richer queries by designing oblivious versions of simple data structures (e.g., maps, sets, queues). In this paper, we take a distinct step by introducing two building blocks that make it easy to construct many advanced oblivious and compressed data structures efficiently and securely. Our approach builds on recent results showing that most (compressed) data structures can be implemented using two fundamental primitives: Rank and Select, which count and locate symbol occurrences within a sequence. We present the first oblivious implementations of Rank and Select that require only a single ORAM access per operation, use O(1) client-side space, and add negligible client computation overhead. Thanks to their simplicity and efficiency, our oblivious Rank and Select provide a foundation for building a wide range of other oblivious compressed data structures, enabling privacy-preserving (compressed) data storage, indexing, and search in the cloud.
Download

Short Papers

Paper Nr:	173
Title:	Traphalt: Indirect Core Halt via Trap Handling to Halt a RISC-V Core from User Mode through Kernel-Mediated Fault Handling
Authors:	Muhammad Awais, Maria Mushtaq, Lirida Naviner, Jawad Haj Yahya and Florent Bruguier
Abstract:	Modern processors rely on strict privilege separation to prevent user applications from manipulating privileged architectural state directly. In RISC-V systems, user-mode programs cannot access privileged control and status registers (CSRs) or execute privileged instructions. Still, exceptions raised by user programs are handled by the operating system kernel through the trap mechanism, which runs at a higher privilege level. This interaction between user-level faults and kernel trap handling can introduce subtle system behaviours that affect processor control state. In this work, we show that a RISC-V CPU core can be indirectly halted from user mode by deliberately triggering kernel trap handling with carefully crafted illegal instruction sequences. A user-space program forces the kernel to process a fault condition that propagates through OpenSBI’s unhandled-fault path to sbi hart hang(), suspending the hart in an infinite wfi() loop. We demonstrate this behaviour on a SiFive HiFive Premier P550 platform running Linux with OpenSBI v1.6. Our results highlight how user-triggered exceptions can cause unexpected system-level effects and stress the need for careful validation of exception-handling paths in RISC-V platforms.
Download

Paper Nr:	221
Title:	Dynamic Inertial Sensing for Device Identification
Authors:	Michal Borowski and Gareth Howells
Abstract:	Previous studies have demonstrated that inherent imperfections in inertial sensors can serve as unique identifiers for devices, supporting secure key generation and device authentication. However, these works have primarily focused on measurements taken under static conditions, where sensors were at rest. This paper extends that line of research by investigating inertial sensor-based device identification under dynamic conditions. Specifically, when devices are in motion within a vehicle. We demonstrate that despite motion-induced variations, consistent inter-sensor relationships can be exploited to accurately identify individual devices. Furthermore, we present an ICMetric key generation approach that leverages accelerometer and gyroscope differential readings, using derived alignment and granularity parameters to dynamically regenerate cryptographic keys without storing them. Using five Raspberry Pi 5 devices, each with four MPU-9250 sensors, our experiments demonstrate effective device identification and assess the resilience of motion-regenerated ICMetric-based keys to brute-force attacks.
Download