Security of AI and LLM systems against prompt injection attacks: risk analysis, attack vectors, impacts on confidentiality and integrity, hardening best practices, and defense strategies for applications based on generative models.