DeepTeam is a framework to red team LLMs and AI agents.
import pytest
from deepteam.vulnerabilities import Competition
from deepteam.vulnerabilities.competition import CompetitionType
from deepteam.test_case import RTTestCase
class TestCompetition:
def test_competition_all_types(self):
types = [
"competitor_mention",
"market_manipulation",
"discreditation",
"confidential_strategies",
]
competition = Competition(types=types)
assert sorted(type.value for type in competition.types) == sorted(types)
def test_competition_all_types_default(self):
competition = Competition()
assert sorted(type.value for type in competition.types) == sorted(
type.value for type in CompetitionType
)
def test_competition_competitor_mention(self):
types = ["competitor_mention"]
competition = Competition(types=types)
assert sorted(type.value for type in competition.types) == sorted(types)
def test_competition_market_manipulation(self):
types = ["market_manipulation"]
competition = Competition(types=types)
assert sorted(type.value for type in competition.types) == sorted(types)
def test_competition_discreditation(self):
types = ["discreditation"]
competition = Competition(types=types)
assert sorted(type.value for type in competition.types) == sorted(types)
def test_competition_confidential_strategies(self):
types = ["confidential_strategies"]
competition = Competition(types=types)
assert sorted(type.value for type in competition.types) == sorted(types)
def test_competition_all_types_invalid(self):
types = [
"competitor_mention",
"market_manipulation",
"discreditation",
"confidential_strategies",
"invalid",
]
with pytest.raises(ValueError):
Competition(types=types)
def test_simulate_attacks_returns_expected_cases(self):
competition = Competition(types=["competitor_mention"])
test_cases = competition.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 == "Competition" for tc in test_cases)
assert all(
tc.vulnerability_type == CompetitionType.COMPETITOR_MENTION
for tc in test_cases
)
def test_assess_returns_results(self):
competition = Competition(
types=["competitor_mention"], 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 = competition.assess(
model_callback=dummy_model_callback,
)
assert competition.is_vulnerable() is not None
assert competition.simulated_attacks is not None and isinstance(
competition.simulated_attacks, dict
)
assert competition.res is not None and isinstance(competition.res, dict)
assert CompetitionType.COMPETITOR_MENTION in results
assert len(results[CompetitionType.COMPETITOR_MENTION]) == 1
test_case = results[CompetitionType.COMPETITOR_MENTION][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_Competition_metric(self):
from deepteam.metrics import CompetitorsMetric
competition = Competition(
async_mode=True, verbose_mode=True, evaluation_model="gpt-4o"
)
metric = competition._get_metric(CompetitionType.COMPETITOR_MENTION)
assert isinstance(metric, CompetitorsMetric)
assert metric.async_mode is True
assert metric.verbose_mode is True
@pytest.mark.asyncio
async def test_a_assess_returns_async_results(self):
competition = Competition(types=["competitor_mention"], async_mode=True)
async def dummy_model_callback(prompt):
return prompt
results = await competition.a_assess(
model_callback=dummy_model_callback,
)
assert competition.is_vulnerable() is not None
assert competition.simulated_attacks is not None and isinstance(
competition.simulated_attacks, dict
)
assert competition.res is not None and isinstance(competition.res, dict)
assert CompetitionType.COMPETITOR_MENTION in results
assert len(results[CompetitionType.COMPETITOR_MENTION]) == 1
test_case = results[CompetitionType.COMPETITOR_MENTION][0]
assert hasattr(test_case, "score")
assert hasattr(test_case, "reason")