Adversarially Resilient UX in AI‑Powered Cybersecurity Decision Support
TL;DR
This paper explores the concept of adversarially resilient user experience (UX) in AI-powered cybersecurity tools, and proposes a design framework for resilient UX in such systems through threat modeling, scenario analysis, and a taxonomy of attack vectors at the human-AI interface.
Abstract
As AI becomes integral to cybersecurity decision support systems, the attack surface expands beyond models to include user interfaces and decision workflows. This paper explores the concept of adversarially resilient user experience (UX) in AI-powered cybersecurity tools. We examine how adversaries may exploit cognitive, perceptual, and interaction-level vulnerabilities in UX to mislead human analysts or distort model outputs. Drawing on human-computer interaction (HCI), adversarial ML, and cyber threat intelligence, we propose a design framework for resilient UX in such systems. Through threat modeling, scenario analysis, and a taxonomy of attack vectors at the human-AI interface, we provide guidelines for creating decision-support environments that enhance trust, robustness, and interpretability while resisting manipulation. We validate the approach with case studies in phishing detection and network anomaly triage, and we identify key research directions for building human-AI systems that are secure not just technically, but behaviorally and perceptually as well.