Specific threats and vulnerabilities of generative artificial intelligence systems, including prompt injection, data poisoning, model inversion, extraction of sensitive data from the training set, jailbreak of guardrails, generation of harmful content, amplified biases, and supply chain risks in pre-trained models. Covers LLMs, diffusion models, multimodal generators, and related attack surfaces according to OWASP Top 10 for LLM and NIST AI RMF frameworks.