世界经济论坛：2024年生成人工智能时代的治理报告：一种360°弹性政策和监管方法

Generative AI models amplify privacy, safety and security risks due to their reliance on vast amounts of training data, powerful inference capability and susceptibility to unique adversarial attacks that can undermine digital trust.6 A number of risks arise from the inclusion of personal, sensitive and confidential information in training datasets and user inputs, lack of transparency over the lawful basis for collecting and processing data, the ability of models to infer personal d