COMPARATIVE STUDY OF CYBERSECURITY AI MODELS: PERFORMANCE, SAFETY, TRUST, AND EXPLAINABILITY

The rapid integration of artificial intelligence (AI) in cybersecurity has significantly enhanced threat detection, response automation, and overall network defense (Smith & Zhang, 2021; Patel et al., 2022). However, the effectiveness of AI-driven cybersecurity systems depends not only on technical performance but also on safety, trustworthiness, and explainability (Jones et al., 2020). This study presents a comparative analysis of contemporary AI models applied to cybersecurity, focusing on intrusion detection systems (IDS), malware classification, and Security Operations Center (SOC) automation (Lee & Kim, 2021). Using a curated set of benchmark datasets and realistic threat scenarios, we evaluate each model’s detection performance, vulnerabilities to adversarial attacks, and resilience against model drift (Nguyen et al., 2022). Additionally, we assess trust and explainability metrics, emphasizing human-in-the-loop (HITL) and human-on-the-loop (HOTL) frameworks, and the role of explainable AI (XAI) techniques in improving analyst confidence and decision-making (Doshi-Velez & Kim, 2017; Miller, 2019). Our results highlight critical trade-offs between accuracy, operational safety, and interpretability, revealing that high-performing models may exhibit lower explainability or increased susceptibility to certain adversarial threats (Goodfellow et al., 2015). The study provides a structured framework for selecting AI models in cybersecurity applications, balancing technical efficiency with safety and trust requirements, and offers recommendations for integrating multi-criteria evaluation in real-world deployments (Chio & Freeman, 2018).