First this paper measures and analyzes various trade-offs between different DNSSEC configurations, such as, NSEC, NSEC3, aggressive caching, and online vs. offline signing. These measurements expose a trade-off between (1) aggressive caching DNSSEC configuration that provides excellent performance (higher robustness to DDoS attacks) but is vulnerable to zone enumeration attacks in addition to offline signing disadvantages, and (2) online signing DNSSEC configuration, that prevent zone enumeration but is more susceptible to DDoS attacks due to lower maximum throughput. Second, following these, we suggest and evaluate an alternative, Adaptive DNS over QUIC, (AdaDoQ), an adaptive, efficient and secure communication layer between a DNS resolver and the authoritative servers, that offers a better trade-off between DDoS resiliency and security. Under normal load conditions the resolver communicates with each authoritative server with standard DNSSEC however, when the traffic load increases, AdaDoQ switches to a QUIC connection with heavily communicating authoritative server(s), for as long as the traffic load is high. The public key of the authoritative ZSK (which is verifiable through the DNSSEC chain of trust) is integrated into the symmetric key creation in QUIC to provide a DNSSEC level of authenticity and security. AdaDoQ ensures DNS authenticity, good throughput, and disables zone walking attacks, DNS hijacking, and cache poisoning.
The widespread sharing of anonymized graph datasets for research and public use has raised new privacy concerns, as sensitive information can often be re-identified through advanced deanonymization techniques. Existing graph deanonymization methods span a variety of strategies, including structural alignment, embedding-based matching, and feature extraction combined with machine learning classifiers. However, these approaches remain grounded in graph-specific workflows. In this paper, we introduce G2TA (Graph to Table Attack), a novel attack framework that bridges graph and tabular privacy domains. By transforming anonymized graphs into semantically enriched tabular representations that capture structural patterns and neighborhood-level attribute signals, our approach enables the use of record linkage and quasi-identifier-based attacks originally developed for tabular datasets. These semantic features preserve node-to-neighborhood relationships that are crucial for identity recovery. We evaluate the framework on anonymized graphs and compare it to well-known graph-based attacks, showing that meaningful re-identification is possible, even without graph-specific algorithms. We also evaluated our methodology on differentially private graphs. Our findings underscore the importance of privacy evaluations that account for both graph-specific threats and relational tabular attacks before public release.
Industry 4.0 drives the ICS and SCADA systems to utilize the cloud for its significant benefits. New models, designs, and applications of traditional ICS systems incorporate the cloud to enhance and streamline industrial processes. In this paper, we propose a novel model that utilizes avatar processes for devices in the ICS system, such as PLCs. The avatar runs on the cloud and executes PLC tasks. The avatar can perform heavy computations that were previously impossible, especially for legacy devices standard in these systems. We also describe a security architecture for our system that thwarts cybersecurity threats that may arise from attackers on the cloud. Furthermore, we introduce a new decentralized architecture that combines blockchain with avatars to secure our ecosystem. In total, our system enhances the industrial process and enables engineers to easily incorporate complex logic into their (potentially heterogeneous and legacy) systems.
Managing the timing and delays of computing events is central to distributed systems. Timed cryptography, such as time-lock puzzles (introduced almost 30 years ago) and verifiable delay functions, provides central tools for such tasks: A single party (the committer) knows a secret, commits to it publicly, and then the committed secret can be opened via a time-consuming staged computation by any observer; this process is independent of post-commitment actions by or presence of the committer. Implicit in this scenario is the requirement that the committer itself does not attack the system, and, say, prematurely selectively leak or de-commit the value, either maliciously or due to external attacks. To prevent the above attack and to expand timed-delayed cryptography to address more applications, we initiate here the study of how to distributedly hold a timed secret known neither to any individual party nor to any group smaller than a quorum. This primitive is defined in the context of multi-party computation (MPC). In fact, this implements a setting where a group of parties simultaneously learn a secret that has been kept private for a predetermined amount of time, and could have been collectively generated without a single party controlling its generation (as in distributed key generation scenarios). This new paradigm, and the ability to compose such protocols, also extends time-lock functionalities to applications where a firm commitment by a group of parties to future release is executed even without knowing the underlying secrets and even without assuming cooperation or availability of parties in the future. Sample applications requiring such delayed disclosure include accountable future scanning of software for zero-day vulnerabilities and accountable, secure sharing of clinical trial documents. We design the first secure practical protocol for Multi-party Time-lock Puzzles (MTP), a new primitive centered on the inherently needed distributed generation of a time-lock puzzle, and the inability to prematurely reconstruct the committed value unless a quorum of parties is compromised. We then leverage MTP to build Timed Multi-party Computation (TMPC), which we model as MPC with a single time-locked output.
This paper studies Locally Maximal Satisfying Truth Assignments (LMSTAs) of satisfiable random k-SAT formulas via Kolmogorov complexity. For Kolmogorov-random instances, we show that the fraction of zeros in any LMSTA lies in a contiguous sub-interval of (0, 1) whose bounds depend on the clause-to-variable ratio r. We further derive a mixed-width version that applies to heterogeneous CNFs. As case studies, we analyse the standard SAT encodings of 200-round Trivium and the public 30-round Bivium benchmark and prove that no LMSTAs exist for either, ruling out single-bit “local traps”. This work links information-theoretic randomness to local maximality providing a heuristic, compression-based test for real-world SAT reductions in cryptography.
Range counting is the problem of preprocessing a set P ⊂ℝ^d of n points, such that given a query range γ we can efficiently compute |P ∩γ | . It was already shown (Kushnir et al. PETS’24) how to efficiently answer a range searching query under FHE using a technique they called Copy-and-Recurse to traverse partition trees. In the related Range emptiness problem the goal is to compute only whether P ∩γ = ∅ . This was shown (in plaintext) to be more efficient. In this paper we improve and extend the results of Kushnir et al. First, for range searching we reduce the overhead term to the optimal O(n), so for example if the ranges are halfspaces in ℝ^d bounded by hyperplanes then range searching can be done with a circuit of size O(t · n^1-1/d + ϵ + n) , where t is the size of the sub-circuit that checks whether a point lies under a hyperplane. Second, we introduce a variation of copy-and-recurse that we call leveled copy-and-recurse. With this we improve traversal of trees such as 1D partition trees and binary trees. Third, we show how to answer range emptiness queries under FHE. We implemented our algorithms and show that our techniques for range emptiness yield a solution that is × 3.6 faster than the previous results for a database of 2^25 points.
Cryptocurrencies can be used merely to transfer value between identities, but many of the more interesting uses of cryptocurrencies require contracts, e.g., “a transfer of X coins from party S to party R is permitted only if conditions A and B hold”. Bitcoin (and related cryptocurrencies) place strict limits on the language in which these conditions can be phrased. In particular, conditions have limited length and may not contain loops. In this work, we show how to augment the Bitcoin scripting language with a single “innocuous” operation to allow us to create “meta conditions” with much more expressive power (e.g., as defined by arbitrarily-sized circuits). We construct a protocol to compile such meta-conditions into a set of (augmented) Bitcoin transactions. We then show how to use this compiler to realize a full “meta-ledger” functionality, which we show is secure in the universal composability framework.
Traditional meet in the middle attacks divide the cipher into two halves, and partially encrypt one half and decrypt the other to search for a collision in the middle. In this paper we show that cryptanalytic techniques such as linear cryptanalysis, scattered linear cryptanalysis and differential cryptanalysis can be used as a third (free) zone in between, and allow to perform the meet in the middle analysis on fewer rounds, thus reduce the meet in the middle complexity. I.e., our new meet in the middle attacks divide the analyzed ciphers into three zones, where the middle zone is non-probabilistically approximated by some cryptanalytic approximation (e.g., linear approximation), and the meet in the middle analysis is performed on the two external zones only. We call the middle zone a free zone or a DMZ as the attack does not actively attack it. We exemplify our attacks using the FEAL cipher, which has relatively “long” non-probabilistic approximations .
Let N = pq be the product of two balanced prime numbers p and q. In 2023, Cotan and Teşeleanu introduced a family of RSA-like cryptosystems based on the key equation ed - k(p^n - 1)(q^n - 1) = 1 , where n ≥ 1 . Note that when n = 1 , we obtain the classical RSA system, while n = 2 yields the variant proposed by Elkamchouchi, Elshenawy, and Shaban. In this paper, we present a novel attack that combines continued fractions with lattice-based methods for the case n = 6 . This represents a natural continuation of previous research, which successfully applied similar techniques for n = 1, 2, 4 .
Adversarial patch attacks threaten the reliability of object detectors by causing severe misclassifications, especially in safety-critical environments. In this work, we propose a comprehensive defense pipeline that not only restores detection performance but also significantly improves it. Our method leverages a latent diffusion model to recover semantically coherent regions affected by adversarial patches, leading to confidence gains of +26.61
We examine well known facility location problems under the privacy challenges posed by big data environments. For a given set of n points U∈ℝ^d , previous works have introduced the “Topology Descriptor Grid” (TDG) [13, 14], a privacy-preserving framework under which some approximate solutions are possible for a variety of clustering problems. In this paper, we introduce the Equidistant “Location Estimation using Concentric Circles” (LECC) framework in ℝ^2 , which obfuscates exact point locations while preserving their relative distances to a predetermined point. We show, under this new framework, how to obtain 2+𝒪(1/n) -approximate solutions for the 1-center, 1-median, 1-mean, and k-centrum problems, and 𝒪(k),𝒪(k),𝒪(k^2) approximations for the k-center, k-median and k-means problems, respectively. For the TDG framework we provide a (√(d),k^d-1),(d,k^d-1) , and (d^2,k^d-1) approximations for the k-center, k-median, and k-means problems, respectively.
The increasing adoption of macOS devices leads to a rise in malware targeting the platform, while the current research state of macOS malware detection remains limited compared to Windows. This study aims to enhance macOS malware detection using supervised machine learning techniques. Unlike previous studies, this study integrates multiple feature selection methods, applies nested cross-validation, and introduces a novel RGB image conversion of structured features combined with CycleGAN-generated data for Convolutional Neural Network (CNN) training. The CyberScienceLab macOS malware dataset is used, where comprehensive data preprocessing, feature selection, and balancing through SMOTE are applied to train five supervised machine learning models: Support Vector Machine (SVM), Decision Tree, Naïve Bayes, Random Forest and Logistic Regression. Each model undergoes nested cross-validation and hyperparameter tuning to ensure robust evaluation. For CNN training, RGB images are generated by transforming structural static feature vectors. CycleGAN is employed to generate synthetic image data to address the small dataset size. The results show that all supervised learning models outperform previous studies using the same dataset, with SVM achieving the highest accuracy of 98.88
The problem of similarity-based retrieval, in which a server retrieves a vector from a database that is most similar to a query of a client, is a fundamental problem for many applications. Fully Homomorphic Encryption (FHE) supports computations over encrypted data and thus can be used to preserve the privacy of the query and database during the similarity-based retrieval process. However, existing works that tackle the problem of similarity-based retrieval over FHE typically rely on sending an encrypted vector containing several computed similarity scores from the server to the client. This client-aided approach exposes too much information on the dataset, while also incurring high communication bandwidth that is linear in the size of the dataset. In this work, we present a similarity-based retrieval system in which the server sends the client only one ciphertext containing the retrieved entry, thus not exposing additional information on the dataset while improving the communication bandwidth to be constant. We conduct empirical experiments in which we perform similarity-based retrieval over a dataset of half a million encrypted vectors in less than 30 s with accuracy of 98.9 8 × in the case of a single query vector and by 512 × in the case of a batch of 64 query vectors.
Path-aware Internet architectures such as SCION expose multiple end-to-end paths to endhosts. However, optimal path selection in dynamic network conditions remains challenging. This paper presents a Deep-Q-Network approach for intelligent path selection in SCION networks. We formulate SCION path selection as a reinforcement learning problem and train a lightweight Deep-Q-Network agent that observes latency, loss, and bandwidth metrics and outputs the optimal path. We evaluate our approach in an simulated SCION environment with realistic time-varying traffic conditions. The DQN agent consistently matches the performance of oracle-based selection methods with full network visibility, while reducing probing overhead by 95
This study focuses on developing an automated literary criticism (ALC) system using Large Language Models (LLMs). The ALC system aims to analyze the stylistic, structural, and thematic features of short literary texts, particularly movie synopses. By combining quantitative indicators, such as stylistic consistency and consistency of presentation, with qualitative assessments, the system aims to provide authors and researchers with detailed feedback regarding the content. The ALC goal is to reliably reproduce professional critical assessments by evaluating texts across key narrative criteria such as Character, Conflict, Originality, Logic, and Premise. Focusing on movie synopses and other short literary texts offers an ideal starting point due to their short length, availability of external quality benchmarks, and manageable length for consistent LLM evaluation. Our initial results demonstrate that LLMs, without prior domain-specific fine-tuning or calibration, can distinguish between high-quality and low-quality synopses, as evidenced by statistically significant differences between evaluations scores of Oscar-winning screenplays and Golden Raspberry winners. An automated evaluation system providing objective, structured, and evidence-based critiques aligned with human expert assessments can facilitate rapid pre-evaluation of scripts and literary texts in publishing and media industries, enhancing editorial decision-making processes. In the future, we plan to extend the proposed framework to broader literary genres and conduct user studies to evaluate alignment with professional critics.
Detecting fraud within financial systems has become important due to the continuous evolution of fraudulent techniques to evade existing detection methods. Credit card fraud is one of the most common forms of payment fraud and is steadily rising globally. In this work, we proposed a framework that leverages three distinct multi-view graphs, capturing relationships between cardholders and merchants, transaction categories, and geospatial merchant clusters. These are integrated with a temporally causal transaction sequence graph to enhance fraud detection in credit card transactions. To effectively learn meaningful edge representations, the framework employs a Graph Attention Network (GAT) for fusing the heterogeneous features derived from these graph views. We also mitigate the challenge of class imbalance using the Synthetic Minority Oversampling Technique (SMOTE), facilitating two experimental settings, one using an undersampled dataset and the other using an oversampled dataset through bootstrapped minority instances. We perform various experiments, and the proposed model demonstrates significantly high performance compared to traditional models and current state-of-the-art approaches, attaining an accuracy of 88.93
Random sequence construction represents fundamental components of modern cryptographic systems. The quantitative assessment of randomness relies upon rigorous statistical testing methodologies, establishing statistical randomness evaluation as a critical prerequisite for cryptographic algorithm security validation. Concurrently, data compression technologies have emerged as essential enablers of efficient information transmission within contemporary digital communication infrastructures. This research investigates statistical testing frameworks in cryptographic applications, with particular emphasis on compression-based evaluation methods, notably the Lempel-Ziv complexity test. We present empirical findings from our analysis of a novel bit-level pattern recognition algorithm, validated against data sequences generated through the Advanced Encryption Standard (AES). Furthermore, we introduce a compression methodology derived from this bit-level pattern detection approach, demonstrating its potential applications in cryptographic randomness assessment.
