Explainability-Guided Black-Box Adversarial Auditing for Detecting Hidden Data Poisoning in Multimodal Foundation Models. JDIAS 2026, 1 (3).