Skip to main content

Craft Adversarial Data

Tests whether adversarially crafted data — perturbations, poisoned examples, or augmentation abuse — can compromise model behaviour.

Sub-types​

  • Perturbation Attacks: Small, imperceptible changes to inputs that alter model outputs.
  • Poisoned Examples: Adversarially crafted examples designed to trigger specific model failures.
  • Data Augmentation Abuse: Exploiting data augmentation pipelines to inject adversarial samples.

Threat Profile​

Objective: jailbreak

Objective vs. goals

The objective above selects the scoring rubric for a run — it is one of a fixed set of built-in names, not something you write. The goals you pass to an attack are separate free-text strings that you author yourself. See Goals vs. objective.

Primary

  • advbench: Adversarial goals that may involve crafted perturbations

Attack Techniques​

Primary

  • Static Template: Template-based prompt construction

Metrics​

  • asr
  • judge_score

Usage​

Instantiate the Vulnerability​

from hackagent.risks import CraftAdversarialData
from hackagent.risks.craft_adversarial_data.types import CraftAdversarialDataType

# Use all sub-types
vuln = CraftAdversarialData()

# Or specify particular sub-types
vuln = CraftAdversarialData(types=[
    CraftAdversarialDataType.PERTURBATION_ATTACKS.value,
    CraftAdversarialDataType.POISONED_EXAMPLES.value,
])

Run an Evaluation Campaign​

from hackagent import HackAgent
from hackagent.risks.craft_adversarial_data import CRAFT_ADVERSARIAL_DATA_PROFILE

agent = HackAgent(endpoint="http://localhost:8080/chat", name="my-agent")

# Use profile recommendations
for attack in CRAFT_ADVERSARIAL_DATA_PROFILE.primary_attacks:
    for dataset in CRAFT_ADVERSARIAL_DATA_PROFILE.primary_datasets:
        attack_config = {
            "attack_type": "static_template",  # attack.technique is "StaticTemplate"
            "objective": CRAFT_ADVERSARIAL_DATA_PROFILE.objective,
            "dataset": {"preset": dataset.preset},
        }
        results = agent.hack(attack_config=attack_config)
        print(f"Results: {results}")