Publication Search

The modernization of smart grids through edge computing introduces significant cybersecurity challenges, most stemming from adversarial machine learning attacks that compromise distributed intelligence. Although Federated Learning is an appealing decentralized model training paradigm for edge smart grids, its resilience against coordinated injection and evasion attacks has not yet been thoroughly explored. To address this critical gap, we develop and evaluate a resilient Federated Learning model for edge-based innovative grid applications. Under a rigorous simulation-based experimental design, we created a controlled environment based on synthetic energy-demand data and implemented adversarial attack scenarios to ensure model robustness. We propose a resilience enhancement layer in our framework during the federated aggregation process to curtail malicious model updates and adversarial inferences. The results show significant improvement in the stability of the proposed model under attack, maintaining a robustness index above 0.62, whereas baseline approaches exhibit complete degradation. This corresponds to a reduction of approximately 34% in the attack impact rate across different-intensity attack scenarios, while maintaining high stability in aggregation. In addition to the adversarial testing framework in the domain of Federated Learning, this work provides a validated resilience model that secures analytics of smart grids without requiring access to raw data. Our methodology presents a resource-efficient alternative to physical testing and enables safe yet comprehensive security evaluation in critical infrastructure applications.

The rapid deployment of large language models (LLMs) on mobile devices has introduced significant privacy concerns, particularly regarding data collection, user profiling, and compliance with evolving AI regulations such as the GDPR and the AI Act. While these on-device LLMs promise improved latency and user experience, their potential to inadvertently leak sensitive information remains understudied. This paper proposes a red teaming framework to systematically assess the privacy risks of phone-based LLMs, simulating adversarial attacks to identify vulnerabilities in model behavior, data storage, and inference processes. We evaluate popular mobile LLMs under scenarios such as prompt injection, side-channel exploitation, and unintended memorization, measuring their compliance with strict privacy-by-design principles. Our findings reveal critical gaps in current safeguards, including susceptibility to context-aware deanonymization and insufficient data minimization. We further discuss regulatory implications, advocating for adaptive red teaming as a mandatory evaluation step in AI governance. By integrating adversarial testing into the development lifecycle, stakeholders can preemptively align phone-based AI systems with legal and ethical privacy standards while maintaining functional utility.