As AI agents act across systems, firewalls reveal what they reach, enforce policy at boundaries, and help contain the risks of autonomous action.