GenAI agents can write, click, and execute, but without the right guardrails, they can also leak, inject, and attack.