Exploiting Vulnerabilities and Security Threats in Retrieval-Augmented Generative Models: The LIAR Attack Framework