DeepTeam is a framework to red team LLMs and AI agents.
import pytest
from deepteam.frameworks import NIST
from deepteam.red_teamer.risk_assessment import RiskAssessment
from deepteam.vulnerabilities import BaseVulnerability
from deepteam.attacks import BaseAttack
from deepteam.frameworks.nist.risk_categories import NIST_CATEGORIES
from deepteam.frameworks.risk_category import RiskCategory
from deepteam import red_team
class TestNIST:
def test_nist_init(self):
"""Test that NIST framework can be instantiated."""
framework = NIST()
assert framework is not None
def test_nist_name(self):
"""Test that NIST framework has correct name."""
framework = NIST()
assert (
framework.name
== framework.get_name()
== "NIST AI Risk Management Framework (AI RMF)"
)
def test_nist_description(self):
"""Test that NIST framework has correct description."""
framework = NIST()
assert (
framework.description
== "A structured methodology from NIST for identifying, evaluating, and mitigating risks in AI systems. Confident AI's implementation focuses on the Measure function - testing, evaluation, and assurance of AI behavior and risk controls across four measurement categories (M.1-M.4) covering risk metrics, trustworthiness evaluation, risk tracking, and impact assessment."
)
def test_nist_default_categories(self):
"""Test that all four measure categories are included by default."""
framework = NIST()
assert set(framework.categories) == {
"measure_1",
"measure_2",
"measure_3",
"measure_4",
}
def test_nist_partial_categories(self):
"""Test that framework can be created with limited categories."""
framework = NIST(categories=["measure_1", "measure_2"])
assert set(framework.categories) == {"measure_1", "measure_2"}
def test_nist_vulnerabilities_exist(self):
"""Test that vulnerabilities are defined and populated."""
for risk_category in NIST_CATEGORIES:
assert hasattr(risk_category, "vulnerabilities")
assert risk_category.vulnerabilities is not None
assert len(risk_category.vulnerabilities) > 0
def test_nist_vulnerabilities_are_instances(self):
"""Test that all vulnerabilities are instances of BaseVulnerability."""
for risk_category in NIST_CATEGORIES:
for vuln in risk_category.vulnerabilities:
assert isinstance(vuln, BaseVulnerability)
def test_nist_attacks_exist(self):
"""Test that attacks are defined and populated."""
for risk_category in NIST_CATEGORIES:
assert hasattr(risk_category, "attacks")
assert risk_category.attacks is not None
assert len(risk_category.attacks) > 0
def test_nist_attacks_are_instances(self):
"""Test that all attacks are instances of BaseAttack."""
for risk_category in NIST_CATEGORIES:
for attack in risk_category.attacks:
assert isinstance(attack, BaseAttack)
def test_nist_vulnerability_types_present(self):
"""Test that key vulnerabilities are present."""
vuln_names = []
for risk_category in NIST_CATEGORIES:
vuln_names.extend(
[v.__class__.__name__ for v in risk_category.vulnerabilities]
)
expected_vulns = [
"IntellectualProperty",
"RBAC",
"Bias",
"Fairness",
"Ethics",
"Toxicity",
"Misinformation",
"PIILeakage",
"PromptLeakage",
"IllegalActivity",
"Robustness",
"ExcessiveAgency",
"ShellInjection",
"SQLInjection",
"SSRF",
"GraphicContent",
"PersonalSafety",
"ChildProtection",
"BOLA",
"BFLA",
"Competition",
]
for name in expected_vulns:
assert name in vuln_names, f"Missing vulnerability {name}"
def test_nist_attack_types_present(self):
"""Test that key attacks are included."""
attack_names = []
for risk_category in NIST_CATEGORIES:
attack_names.extend(
[a.__class__.__name__ for a in risk_category.attacks]
)
expected_attacks = [
"PromptInjection",
"PromptProbing",
"Base64",
"ROT13",
"Leetspeak",
"Multilingual",
"GrayBox",
"Roleplay",
"CrescendoJailbreaking",
"LinearJailbreaking",
"TreeJailbreaking",
"SequentialJailbreak",
"BadLikertJudge",
]
for attack in expected_attacks:
assert attack in attack_names, f"Expected attack {attack} not found"
def test_nist_attack_weights_defined(self):
"""Test that all attacks have a valid weight."""
for risk_category in NIST_CATEGORIES:
for attack in risk_category.attacks:
assert hasattr(attack, "weight")
assert isinstance(attack.weight, int)
assert attack.weight >= 1
def test_nist_category_vulnerability_mapping(self):
"""Test that all categories map to vulnerabilities properly."""
categories = NIST_CATEGORIES
assert set([category.name for category in categories]) == {
"measure_1",
"measure_2",
"measure_3",
"measure_4",
}
for risk_category in categories:
assert isinstance(risk_category, RiskCategory)
assert all(
isinstance(v, BaseVulnerability)
for v in risk_category.vulnerabilities
)
def test_nist_category_attack_mapping(self):
"""Test that all categories map to attacks properly."""
categories = NIST_CATEGORIES
assert set([category.name for category in categories]) == {
"measure_1",
"measure_2",
"measure_3",
"measure_4",
}
for risk_category in categories:
assert isinstance(risk_category, RiskCategory)
assert all(isinstance(v, BaseAttack) for v in risk_category.attacks)
def test_nist_framework_with_red_team(self):
import random
"""Test that NIST framework integrates correctly with red_team function."""
# Simple mock model for testing
def mock_model_callback(prompt: str, turns=None) -> str:
return (
"I'm a safe AI model. I cannot comply with unsafe instructions."
)
categories = ["measure_1", "measure_2", "measure_3", "measure_4"]
random_category = random.choice(
categories
) # Randomizing to not test on too many attacks or vulnerabilities
risk_assessment = red_team(
model_callback=mock_model_callback,
framework=NIST(categories=[random_category]),
async_mode=False,
ignore_errors=False,
)
assert isinstance(risk_assessment, RiskAssessment)
assert risk_assessment is not None