to others through the outputs generated by these models. Data collection can occur not only during the model development and training phase but also at the deployment and use stage, such as the prompts and queries that users enter into services such as ChatGPT. In such cases, not only personal data but also sensitive work-related materials may become subject to security risks. Moreover, as seen in examples such as Copilot integrated into the Windows operating system or Meta’s AI-powered glasses, data collection is increasingly extending beyond specific web services to encompass users’ entire computing environments and everyday physical spaces. As these points of data collection expand, so too do the potential sites of security vulnerability. Generative AI Guide for Civil Society

Select target paragraph3