DeepTeam is a framework to red team LLMs and AI agents.
from deepteam import red_team
from deepteam.vulnerabilities import Bias
from deepteam.attacks.single_turn import *
from deepteam.attacks.multi_turn import *
async def model_callback(input: str, turns=None) -> str:
return input
def test_red_teaming():
vulnerabilities = [Bias(types=["race"])]
attacks = [
# Single-turn attacks
AdversarialPoetry(),
Base64(),
GrayBox(),
Leetspeak(),
MathProblem(),
Multilingual(),
PromptInjection(),
PromptProbing(),
Roleplay(
role="learned gentleman of natural philosophy",
persona="1600s Shakespearean scholar",
),
ROT13(),
SystemOverride(),
PermissionEscalation(),
LinguisticConfusion(),
InputBypass(),
ContextPoisoning(),
GoalRedirection(),
# Multi-turn attacks
CrescendoJailbreaking(),
LinearJailbreaking(),
TreeJailbreaking(),
SequentialJailbreak(),
BadLikertJudge(),
]
risk_assessment = red_team(
model_callback=model_callback,
vulnerabilities=vulnerabilities,
attacks=attacks,
attacks_per_vulnerability_type=1,
max_concurrent=20,
ignore_errors=False,
)
# assert len(risk_assessment.test_cases) > len(vulnerabilities) * len(attacks)
assert risk_assessment is not None
assert len(risk_assessment.test_cases) == 1