Three separate security disclosures this week exposed a pattern: we are deploying agentic AI infrastructure faster than we can secure it, from MCP servers to coding assistants.