I want to understand whether generative AI systems can leak sensitive or private information. How can AI models accidentally expose training data, personal details, or confidential content? Can someone also explain the risks and best practices to prevent such leaks?