Skip to main content

hackagent.attacks.techniques.baseline.attack

Baseline attack implementation.

Sends goals directly to the target model without any transformation, serving as a control condition for measuring default refusal rates.

BaselineAttack Objects​

class BaselineAttack(BaseAttack)

Baseline attack that sends goals directly to the target.

No prompt transformation is applied — goals are sent as-is. This provides a control condition to compare against actual attack techniques (PAIR, TAP, DrAttack, etc.).

Pipeline stages​

  1. Generation — sends each goal verbatim to the target model.
  2. Evaluation — scores responses using the configured evaluator.

get_effective_model_roles​

@classmethod
def get_effective_model_roles(
    cls,
    attack_config: Dict[str, Any],
    *,
    goal_labels_by_index: Optional[Dict[int, Dict[str, str]]] = None
) -> List[Dict[str, Any]]

Baseline always needs judge models for LLM-judge evaluation.

run​

@with_tui_logging(logger_name="hackagent.attacks", level=logging.INFO)
def run(goals: List[str]) -> Dict[str, Any]

Execute baseline attack (direct goal submission).

Arguments:

  • goals - List of goal strings to send directly.

Returns:

Dictionary with 'evaluated' and 'summary' DataFrames.