Skip to content

DeepTeam is a framework to red team LLMs and AI agents.

tests updated

A-Vamshi6f613490c62b
diff --git a/tests/test_frameworks/test_aegis.py b/tests/test_frameworks/test_aegis.py
index 1c324425..cfc04d91 100644
--- a/tests/test_frameworks/test_aegis.py
+++ b/tests/test_frameworks/test_aegis.py
@@ -20,7 +20,7 @@ class TestAegis:
framework = Aegis()
assert (
framework.description
- == "Nemotron Content Safety Dataset V1, formerly known as Aegis AI Content Safety Dataset, is an open-source content safety dataset (CC-BY-4.0), which adheres to Nvidia's content safety taxonomy, covering 13 critical risk categories"
+ == "A dataset-driven safety framework that uses the NVIDIA Aegis AI Content Safety Dataset — an open-source dataset aligned with NVIDIA's Content Safety Taxonomy. Tests models using real human-labeled safety violations from public conversations across 13 critical harm categories including sexual content, violence, hate speech, self-harm, misinformation, privacy violations, child exploitation, drugs, and terrorism. Enables evaluation of model refusal behaviors and fine-tuning of safety filters using authentic, labeled harm categories."
)
def test_aegis_red_team(self):
diff --git a/tests/test_frameworks/test_beavertails.py b/tests/test_frameworks/test_beavertails.py
index 5ee3f914..55192aad 100644
--- a/tests/test_frameworks/test_beavertails.py
+++ b/tests/test_frameworks/test_beavertails.py
@@ -20,7 +20,7 @@ class TestBeaverTails:
framework = BeaverTails()
assert (
framework.description
- == "BeaverTails is an AI safety-focused collection comprising a series of datasets, it contains human-labeled data consisting of question-answer (QA) pairs from 14 different harm categories."
+ == "A dataset-driven safety framework that uses the BeaverTails dataset — a large, human-labeled collection of harmful and borderline prompts. Tests models against real-world adversarial inputs across 14 harm categories including abuse, illegal instructions, privacy violations, self-harm, sexual content, and misinformation. Provides higher-fidelity signals by sampling actual annotated malicious prompts rather than synthetic generation, helping validate safety defenses and content filters."
)
def test_beavertails_red_team(self):
diff --git a/tests/test_frameworks/test_mitre.py b/tests/test_frameworks/test_mitre.py
index 7a2fda19..281bb1aa 100644
--- a/tests/test_frameworks/test_mitre.py
+++ b/tests/test_frameworks/test_mitre.py
@@ -25,7 +25,7 @@ class TestMITRE:
framework = MITRE()
assert (
framework.description
- == "MITRE ATLAS mapping for LLM/ML adversarial techniques (project-specific) to understand and defend against threats specific to AI and ML systems."
+ == "A structured knowledge base of adversarial tactics, techniques, and procedures (TTPs) used against AI and ML systems. Extends MITRE ATT&CK® principles to the AI threat surface, testing across six adversarial tactics: reconnaissance (information gathering), resource development (capability building), initial access (entry point exploitation), ML attack staging (model-specific preparation), exfiltration (data/model theft), and impact (manipulation/degradation)."
)
def test_mitre_default_categories(self):
diff --git a/tests/test_frameworks/test_nist.py b/tests/test_frameworks/test_nist.py
index 7d24074c..a93c6382 100644
--- a/tests/test_frameworks/test_nist.py
+++ b/tests/test_frameworks/test_nist.py
@@ -25,7 +25,7 @@ class TestNIST:
framework = NIST()
assert (
framework.description
- == "NIST AI Risk Management Framework (AI RMF) — Measure-focused mapping for testing and evaluation."
+ == "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):
diff --git a/tests/test_frameworks/test_owasp_top_10.py b/tests/test_frameworks/test_owasp_top_10.py
index 02ad7066..d730f23a 100644
--- a/tests/test_frameworks/test_owasp_top_10.py
+++ b/tests/test_frameworks/test_owasp_top_10.py
@@ -23,7 +23,10 @@ class TestOWASP:
def test_owasp_description(self):
"""Test that OWASP framework has correct description."""
framework = OWASPTop10()
- assert framework.description == "The OWASP Top 10 for LLMs 2025"
+ assert (
+ framework.description
+ == "A comprehensive list of the most critical security risks associated with LLM applications. The 2025 edition includes 10 critical risks covering prompt injection, sensitive information disclosure, supply chain vulnerabilities, data poisoning, output handling, excessive agency, system prompt leakage, vector/embedding weaknesses, misinformation, and unbounded consumption. Each risk is tested using realistic attack techniques and vulnerability assessments."
+ )
def test_owasp_default_categories(self):
"""Test that all 10 OWASP LLM categories are included by default."""