Users should be told when they are interacting with AI and understand the system's limitations. Which principle most directly supports this requirement?

1. Availability-zone redundancy
2. Transparency
3. Resource autoscaling
4. Encryption alone