DeepTeam is a framework to red team LLMs and AI agents.
import pytest
from deepteam.vulnerabilities import Robustness
from deepteam.vulnerabilities.robustness import RobustnessType
from deepteam.test_case import RTTestCase
class TestRobustness:
def test_robustness_all_types(self):
types = ["input_overreliance", "hijacking"]
robustness = Robustness(types=types)
assert sorted(type.value for type in robustness.types) == sorted(types)
def test_robustness_all_types_default(self):
robustness = Robustness()
assert sorted(type.value for type in robustness.types) == sorted(
type.value for type in RobustnessType
)
def test_robustness_input_overreliance(self):
types = ["input_overreliance"]
robustness = Robustness(types=types)
assert sorted(type.value for type in robustness.types) == sorted(types)
def test_robustness_hijacking(self):
types = ["hijacking"]
robustness = Robustness(types=types)
assert sorted(type.value for type in robustness.types) == sorted(types)
def test_robustness_all_types_invalid(self):
types = ["input_overreliance", "hijacking", "invalid"]
with pytest.raises(ValueError):
Robustness(types=types)
def test_simulate_attacks_returns_expected_cases(self):
robustness = Robustness(types=["hijacking"])
test_cases = robustness.simulate_attacks(
attacks_per_vulnerability_type=2
)
assert len(test_cases) == 2
assert all(isinstance(tc, RTTestCase) for tc in test_cases)
assert all(tc.vulnerability == "Robustness" for tc in test_cases)
assert all(
tc.vulnerability_type == RobustnessType.HIJACKING
for tc in test_cases
)
def test_assess_returns_results(self):
robustness = Robustness(types=["hijacking"], async_mode=False)
def dummy_model_callback(prompt):
# Provide a simple pass-through or minimal callback if required by your real env
return prompt
results = robustness.assess(
model_callback=dummy_model_callback,
)
assert robustness.is_vulnerable() is not None
assert robustness.simulated_attacks is not None and isinstance(
robustness.simulated_attacks, dict
)
assert robustness.res is not None and isinstance(robustness.res, dict)
assert RobustnessType.HIJACKING in results
assert len(results[RobustnessType.HIJACKING]) == 1
test_case = results[RobustnessType.HIJACKING][0]
# The score and reason will depend on the real metric evaluation
assert hasattr(test_case, "score")
assert hasattr(test_case, "reason")
def test_get_metric_returns_Robustness_overreliance_metric(self):
from deepteam.metrics import OverrelianceMetric
robustness = Robustness(
async_mode=True, verbose_mode=True, evaluation_model="gpt-4o"
)
metric = robustness._get_metric(RobustnessType.INPUT_OVERRELIANCE)
assert isinstance(metric, OverrelianceMetric)
assert metric.async_mode is True
assert metric.verbose_mode is True
def test_get_metric_returns_Robustness_hijacking_metric(self):
from deepteam.metrics import HijackingMetric
robustness = Robustness(
async_mode=True, verbose_mode=True, evaluation_model="gpt-4o"
)
metric = robustness._get_metric(RobustnessType.HIJACKING)
assert isinstance(metric, HijackingMetric)
assert metric.async_mode is True
assert metric.verbose_mode is True
@pytest.mark.asyncio
async def test_a_assess_returns_async_results(self):
robustness = Robustness(types=["hijacking"], async_mode=True)
async def dummy_model_callback(prompt):
return prompt
results = await robustness.a_assess(
model_callback=dummy_model_callback,
)
assert robustness.is_vulnerable() is not None
assert robustness.simulated_attacks is not None and isinstance(
robustness.simulated_attacks, dict
)
assert robustness.res is not None and isinstance(robustness.res, dict)
assert RobustnessType.HIJACKING in results
assert len(results[RobustnessType.HIJACKING]) == 1
test_case = results[RobustnessType.HIJACKING][0]
assert hasattr(test_case, "score")
assert hasattr(test_case, "reason")