Abstract

Transformer-based Intrusion Detection and Response Systems (T-IDRS) offer sophisticated, autonomous cyber defense, yet their adoption is hindered by the "black-box" nature of machine learning. This opacity often leads analysts to either blindly accept or reject system recommendations, undermining security efficacy. This research evaluates human-centered explanation interfaces designed to align AI outputs with analysts’ mental models. By implementing three distinct frameworks (counterfactual explanations, feature-attribution visualizations, and attack-path maps) the study moves beyond technical explainability toward functional transparency. Through a controlled user study with security analysts, the project plans to measure trust, decision quality, and cognitive workload across multiple simulated attack scenarios. The findings will provide empirical evidence of explanation effectiveness within professional workflows and establish human-centered design guidelines for AI-driven defense. Ultimately, this work contributes a reusable benchmark of interfaces and datasets to enhance the utility and trustworthiness of autonomous security systems.

Download

Share

COinS