From Generative to Agentic AI: What It Means for Knowledge Safety and Cybersecurity