AI agents don't need to break in. That's the real problem.

Explores why AI agents pose a distinct identity and access risk since they operate with granted credentials rather than needing to breach defenses, and argues for treating agent identity as a first-class security concern.