CURE presents a training-free framework for fast, reliable concept unlearning in diffusion models for targeted artistic styles, objects, identities, or explicit content, that is robust against red-teaming attacks.