As AI agents gain autonomy and access to sensitive systems, emerging threats like prompt injection worms highlight how human-like security training and governance must evolve to prevent large-scale, opaque cybersecurity breaches driven by agent behavior.