Risks and countermeasures for systems based on generative models such as LLMs, diffusion models, and multimodal generators. Covers prompt injection, data poisoning, model inversion, jailbreak, hallucination exploitation, sensitive data leakage from the training set, and AI model supply chain governance.