• Make AIs want to whistleblow. Train AIs to report to channels (or eachother) if they suspect they are involved in malicious activity.

This is genius.