Bad Actor Ruleset

Bad Actor is designed to prevent bad actors from interacting with your LLM applications through the detection of jailbreak and injection attacks.

The following yaml code can be added to your policy to enable the Bad Actor ruleset.

  - ruleset: score.bad_actors
    options:
      behavior: observe
      sensitivity: medium

This ruleset adds the equivalent of the following metric section to your yaml policy and uses those metrics to compute an overall score.

metrics:
  - metric: prompt.similarity.jailbreak
  - metric: prompt.similarity.injection