arXiv subjects

Zisis Tsiatsikas

Publications and source records attributed to Zisis Tsiatsikas.

3 recordsLinked to original sources

arXiv2026

Protecting On-Device AI Inference: A Systematic Review of Attacks and Defence Mechanisms

The need for secure and private Artificial Intelligence (AI) and Machine Learning (ML) on edge and mobile devices has increased the necessity of protecting the architecture of these systems from threats to both security and privacy. With an ever-increasing number of pre-trained AI models being used on mobile platforms for client-side inference, there are rising concerns about the risks associated with the theft/extraction of AI models, adversarial attacks on AI models, and data breaches. As a result of this trend, a variety of defence mechanisms have been proposed to protect against these threats. These include Trusted Execution Environments (TEEs), homomorphic encryption, obfuscation, and differential privacy, among others. However, current surveys largely focus on edge intelligence, which includes distributed training, and thus overlook security and privacy issues that are specific to on-device AI inference. To the best of our knowledge, this paper presents the first comprehensive review of threats and corresponding defence mechanisms targeting on-device inference. Our results show that the attack and defence literature are unbalanced: approximately one quarter of the surveyed attack papers focus on Intellectual Property (IP) attacks, whereas half of the defence solutions tackle the same issue. More importantly, some attack categories have no defence paper associated to them, such as adversarial attacks that account for roughly one third of the attack literature. This asymmetry between known attacks and available mitigations highlights clear opportunities for future research on securing on-device AI inference.

cs.CR↗

arXiv2024

Keep your memory dump shut: Unveiling data leaks in password managers

Password management has long been a persistently challenging task. This led to the introduction of password management software, which has been around for at least 25 years in various forms, including desktop and browser-based applications. This work assesses the ability of two dozen password managers, 12 desktop applications, and 12 browser-plugins, to effectively protect the confidentiality of secret credentials in six representative scenarios. Our analysis focuses on the period during which a Password Manager (PM) resides in the RAM. Despite the sensitive nature of these applications, our results show that across all scenarios, only three desktop PM applications and two browser plugins do not store plaintext passwords in the system memory. Oddly enough, at the time of writing, only two vendors recognized the exploit as a vulnerability, reserving CVE-2023-23349, while the rest chose to disregard or underrate the issue.

cs.CR↗

arXiv2023

Bypassing antivirus detection: old-school malware, new tricks

Being on a mushrooming spree since at least 2013, malware can take a large toll on any system. In a perpetual cat-and-mouse chase with defenders, malware writers constantly conjure new methods to hide their code so as to evade detection by security products. In this context, focusing on the MS Windows platform, this work contributes a comprehensive empirical evaluation regarding the detection capacity of popular, off-the-shelf antivirus and endpoint detection and response engines when facing legacy malware obfuscated via more or less uncommon but publicly known methods. Our experiments exploit a blend of seven traditional AV evasion techniques in 16 executables built in C++, Go, and Rust. Furthermore, we conduct an incipient study regarding the ability of the ChatGPT chatbot in assisting threat actors to produce ready-to-use malware. The derived results in terms of detection rate are highly unexpected: approximately half of the 12 tested AV engines were able to detect less than half of the malware variants, four AVs exactly half of the variants, while only two of the rest detected all but one of the variants.

cs.CR↗