The ISC International Journal of Information Security

Abstract Large Language Models (LLMs), trained on extensive datasets using advanced deeplearning architectures, have demonstrated remarkable performance across a wide range of language tasks, becoming a cornerstone of modern AI technologies. However, ensuring their trustworthiness remains a critical challenge, asreliability is essential not only for accurate performance but also for upholding ethical, cultural, and social values. Careful alignment of training data and culturally grounded evaluation criteria is vital for developing responsible AI systems. In this study, we introduce the EPT (Evaluation of Persian Trustworthiness) metric, a culturally informed benchmark specifically designed to assess the trustworthiness of LLMs across six key aspects: Truthfulness, Safety, Fairness, Robustness, privacy, and ethical alignment. We curated a labelled dataset and evaluated the performance of several leading models—including ChatGPT, Claude, DeepSeek, Gemini, Grok, LLaMA, Mistral, and Qwen—using both automated LLM-based and human assessments. Our results reveal significant deficiencies in the safety dimension, underscoring the urgent need for focused attention on this critical aspect of model behaviour. Furthermore, our findings offer valuable insights into the alignment of these models with Persian ethical-cultural values and highlight critical gaps and opportunities for advancing trustworthy and culturally responsible AI. The dataset is publicly available at: https://github.com/Rezamirbagheri110/EPT-Benchmark.

Abstract Recent advances in Machine Learning and Deep Learning have significantly expanded their applications in various domains. The resource-intensive process of training deep neural networks, in terms of substantial labeled data acquisition and computational power, makes these models valuable intellectual property for organizations, hence rising an increasingly crucial need for securing them. A major security threat to deep neural networks is the adversarial examples problem, specifically the black-box type. In these attacks, adversaries generate inputs with often imperceptible crafted perturbations to deceive the model into incorrect classifications, all with no access to the model internals and solely by interacting with it via queries and responses. Among the two primary methods for creating black-box adversarial examples i.e. model extraction-based and query-based approaches, this research focuses on the query-based type, and it explores a novel defense mechanism to mitigate their success. Our proposed method called Divergent Twins Fencing (DTF), employs two subtly different models trained with two different loss functions to incline the execution burden of these attacks. The evaluation criteria for this defense method include measuring the success rate and the average number of queries required to generate adversarial examples using two of the most potent attack methods
from recent studies and comparing its defense performance against a leading defense strategy in the literature, i.e., Random Noise Defense (RND) Method, demonstrating our method’s efficacy in enhancing model security against black-box adversarial attacks.

Abstract While cloud computing is growing at a remarkable speed, privacy issues are far from being solved. One way to diminish privacy concerns is to store data on the cloud in encrypted form. However, encryption often hinders useful computation cloud services. A theoretical approach is to employ the so-called fully homomorphic encryption, yet the overhead is so high that it is not considered a viable solution for practical purposes. The next best thing is to craft special-purpose cryptosystems which support the set of operations required to be addressed by cloud services. In this paper, we put forward one such cryptosystem, which supports efficient search over structured data types, such as timestamps or network addresses, which are comprised of several segments with well-known values. The new cryptosystem, called SESOS, provides the ability to execute LIKE queries, along with the search for exact matches, as well as comparison.
In addition, the extended version, called XSESOS, allows for verifying the integrity of ciphertexts.
At its heart, SESOS combines any order-preserving encryption (OPE) scheme with a novel encryption scheme called Multi-map Perfectly Secure Cryptosystem(MuPS). We prove that MuPS is perfectly secure, and hence SESOS enjoys the same security properties of the underlying OPE scheme.
The overhead of executing equality and comparison operations is negligible. The performance of LIKE queries is significantly improved by up to 1370X and the performance of result decryption improved by 520X compared to existing solutions on a database with merely 100K records (the improvement is even more significant in larger databases).

Abstract One of the security issues in data outsourcing is the enforcement of the data owner’s access control policies. This includes some challenges. The first challenge is preserving confidentiality of data and policies. One of the existing solutions is encrypting data before outsourcing which brings new challenges; namely, the number of keys required to access authorized resources, efficient policy updating, write access control enforcement, overhead of accessing/processing data at the user/owner side. Most of the existing solutions address only some of the challenges, while imposing high overhead on both owner and users. Though, policy management in the Role-Based Access Control (RBAC) model is easier and more efficient due to the existence of role hierarchical structure and role inheritance; most of the existing solutions address only enforcement of policies in the form of access control matrix. In this paper, we propose an approach to enforce RBAC policies on encrypted data outsourced to a service provider. We utilize Chinese Remainder Theorem for key management and role/permission assignment. Efficient user revocation, efficient role hierarchical structure updating, availability of authorized resources for users of new roles, and enforcement of write access control policies as well as static separation of duties, are of advantages of the proposed solution.

Abstract Reputation management systems are in wide-spread use to regulate collaborations in cooperative systems. Collusion is one of the most destructive malicious behaviors in which colluders seek to affect a reputation management system in an unfair manner. Many reputation systems are vulnerable to collusion, and some model-specific mitigation methods are proposed to combat collusion. Detection of colluders is shown to be an NP-complete problem. In this paper, we propose the Colluders Similarity Measure (CSM) which is used by a heuristic clustering algorithm (the Colluders Detection Algorithm (CDA)) to detect colluders in O (n²m + n⁴) in which m and n are the total number of nodes and colluders, respectively. Furthermore, we propose an architecture to implement the algorithm in a distributed manner which can be used together with compatible reputation management systems. Implementation results and comparison with other mitigation methods show that our scheme prevents colluders from unfairly increasing their reputation and decreasing the reputation of the other nodes.

Abstract From the Editor-in-Chief

Abstract From the Editor-in-Chief

Abstract Database outsourcing is an idea to eliminate the burden of database management from organizations. Since data is a critical asset of organizations, preserving its privacy from outside adversary and untrusted server should be warranted. In this paper, we present a distributed scheme based on storing shares of data on different servers and separating indexes from data on a distinct server. Shamir's secret sharing scheme is used for distributing data to data share servers. A B+-tree index on the order preserved encrypted values for each searchable attribute is stored in the index server. To process a query, the client receives responses including record numbers from the index server and asks these records from data share servers. The final result is computed by the client using data shares. While the proposed approach is secure against different database attacks, it supports exact match, range, aggregation, and pattern matching queries efficiently. Simulation results show the prominence of our approach in comparison with the bucketing scheme as it imposes lower computation and communication costs on the client.