Red-team strategy

Disguise and Reconstruction (DRA)

disguise-reconstruction

DRA [Liu 2024]: hides the objective in a bracketed-letter word puzzle, instructs the target to reassemble it in-context, and pins a compliant answer prefix — the request never appears in the prompt, so only output-side attack-success detection can catch it

YAML config

evalguard.config.yaml

redteam:
  strategies:
    - disguise-reconstruction

CLI

bash

# "disguise-reconstruction" is layered onto attacks during red-team scans.
# List every available encoding/obfuscation strategy:
evalguard list strategies

Combine with attack plugins to generate adversarial payloads. The strategy wraps each base payload before it's sent to your target, testing whether the model decodes / unwraps the transform and follows the embedded instruction.