The riskiest agent behavior may not appear in the prompt.
Dangerous Activity Detection in Cortex AI Gateway evaluates the full trajectory: objectives, plans, tool calls, results and effects, with provenance preserved for review.
See how it works:
显示更多