Adversarial Robustness in Women's Safety AI Systems: Threat Taxonomy, ZIDR Metric, and Governance Gap Analysis

This research addresses a documented gap in adversarial robustness evaluation for AI-enabled women's safety systems. Existing evaluation frameworks test these systems under cooperative, benign conditions. No published research examines how a physically proximate adversary - environmentally familiar, non-intimate, adaptive can suppress or corrupt passive detection (computer vision, audio classification, sensor fusion) in the zero-interaction window before an alert is triggered. This deposit contributes: (1) a four-layer threat taxonomy covering sensing, processing, communication, and response attack surfaces; (2) 13 grounded scenarios across urban and rural Indian contexts; (3) governance gap analysis across EU AI Act, NIST AI RMF, ISO 42001, India DPDP Act 2023, and India IT Act; (4) the Zero-Interaction Detection Rate (ZIDR) as a novel evaluation metric for passive-layer robustness under adversarial conditions; and (5) a probe robustness tool specification. Research gap confirmed across nine sources. No existing governance framework requires adversarial robustness testing under physical proximity attack conditions