DeepTeam is a framework to red team LLMs and AI agents.
import pytest
from deepteam.vulnerabilities import BFLA
from deepteam.vulnerabilities.bfla import BFLAType
from deepteam.test_case import RTTestCase
class TestBFLA:
def test_bfla_all_types(self):
types = [
"privilege_escalation",
"function_bypass",
"authorization_bypass",
]
bfla = BFLA(types=types)
assert sorted(type.value for type in bfla.types) == sorted(types)
def test_bfla_all_types_default(self):
bfla = BFLA()
assert sorted(type.value for type in bfla.types) == sorted(
type.value for type in BFLAType
)
def test_bfla_privilege_escalation(self):
types = ["privilege_escalation"]
bfla = BFLA(types=types)
assert sorted(type.value for type in bfla.types) == sorted(types)
def test_bfla_function_bypass(self):
types = ["function_bypass"]
bfla = BFLA(types=types)
assert sorted(type.value for type in bfla.types) == sorted(types)
def test_bfla_authorization_bypass(self):
types = ["authorization_bypass"]
bfla = BFLA(types=types)
assert sorted(type.value for type in bfla.types) == sorted(types)
def test_bfla_all_types_invalid(self):
types = [
"privilege_escalation",
"function_bypass",
"authorization_bypass",
"invalid",
]
with pytest.raises(ValueError):
BFLA(types=types)
def test_simulate_attacks_returns_expected_cases(self):
bfla = BFLA(types=["authorization_bypass"])
test_cases = bfla.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 == "BFLA" for tc in test_cases)
assert all(
tc.vulnerability_type == BFLAType.AUTHORIZATION_BYPASS
for tc in test_cases
)
def test_assess_returns_results(self):
bfla = BFLA(types=["authorization_bypass"], 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 = bfla.assess(
model_callback=dummy_model_callback,
)
assert bfla.is_vulnerable() is not None
assert bfla.simulated_attacks is not None and isinstance(
bfla.simulated_attacks, dict
)
assert bfla.res is not None and isinstance(bfla.res, dict)
assert BFLAType.AUTHORIZATION_BYPASS in results
assert len(results[BFLAType.AUTHORIZATION_BYPASS]) == 1
test_case = results[BFLAType.AUTHORIZATION_BYPASS][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_BFLA_metric(self):
from deepteam.metrics import BFLAMetric
bfla = BFLA(
async_mode=True, verbose_mode=True, evaluation_model="gpt-4o"
)
metric = bfla._get_metric(BFLAType.AUTHORIZATION_BYPASS)
assert isinstance(metric, BFLAMetric)
assert metric.async_mode is True
assert metric.verbose_mode is True
@pytest.mark.asyncio
async def test_a_assess_returns_async_results(self):
bfla = BFLA(types=["authorization_bypass"], async_mode=True)
async def dummy_model_callback(prompt):
return prompt
results = await bfla.a_assess(
model_callback=dummy_model_callback,
)
assert bfla.is_vulnerable() is not None
assert bfla.simulated_attacks is not None and isinstance(
bfla.simulated_attacks, dict
)
assert bfla.res is not None and isinstance(bfla.res, dict)
assert BFLAType.AUTHORIZATION_BYPASS in results
assert len(results[BFLAType.AUTHORIZATION_BYPASS]) == 1
test_case = results[BFLAType.AUTHORIZATION_BYPASS][0]
assert hasattr(test_case, "score")
assert hasattr(test_case, "reason")