Jailbreak

Detect jailbreak and safety-bypass attempts.

YAML key: jailbreak
Direction: input

Detect jailbreak and safety-bypass attempts.

Configuration

Prop

Type

defend.config.yaml (fragment)

guards:
  input:
    modules:
      - jailbreak:
          strict: true

Configure under guards.input.modules when using an LLM-backed input provider.

Injection

Detect prompt injection and instruction-override style attacks.

Invisible text

Flag hidden or unusual Unicode and whitespace abuse often used to evade filters.