Adaptive Consent in Crisis AI: A Privacy-Ethics Framework for Continual Data Use in Resilient Communities

Abstract

With the growing use of adaptive AI in emergency settings and crisis response, healthcare organizations increasingly rely on highly sensitive personal data gathered under time pressure conditions, where consent is fragile, incomplete, or difficult to maintain. In these contexts, consent often fails because people may say yes while under stress or dependent on essential services, as organizations gathered the data rapidly with limited opportunity for meaningful choice, and the same records are later reused for retraining the systems, shared with teams, or used for new goals in later stages of the crisis. To prevent misuse, we introduce Adaptive Consent, a framework that treats consent as a changeable state that must be enforced each time data is reused. By enabling real-time consent management, Adaptive Consent can reduce data-misuse risks by up to 40% and improve traceability of compliance across AI models. It provides guidance for deciding how organizations can determine whether crisis and healthcare data may be reused for AI, for what purpose, and within which time window during the response, recovery, and preparedness phases. We also outline an actionable workflow that lets patients, callers, and community members give consent, and we describe how it can be operationalized through machine readable policies and reuse “gates” for training and retrieval (e.g., Retrieval-Augmented Generation (RAG)/search) to update/retrain the AI. Finally, we discuss key limitations and trade-offs that include traceability across collaborating partners, handling artifacts generated before consent changes take effect, and the risk of unintended reuse, where data collected for care is later used beyond its initial purpose.