Scientists developed an AI monitoring agent to detect and stop harmful outputs

The monitoring system is designed to detect and thwart both prompt injection attacks and edge-case threats.