Skip to content

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

docs update (minor changes)

A-Vamshia0e32b7ffb8c
diff --git a/docs/docs/frameworks-owasp-top-10-for-llms.mdx b/docs/docs/frameworks-owasp-top-10-for-llms.mdx
index 0f49cbc8..712caf2b 100644
--- a/docs/docs/frameworks-owasp-top-10-for-llms.mdx
+++ b/docs/docs/frameworks-owasp-top-10-for-llms.mdx
@@ -62,21 +62,20 @@ The 2025 OWASP Top 10 for LLMs includes significant updates reflecting real-worl
**Indirect Injection:** Hidden instructions in external content (documents, websites, emails) that the LLM processes.
-```python title="main.py"
+```python
+from deepteam.frameworks import OWASPTop10
from deepteam import red_team
-from deepteam.attacks.single_turn import PromptInjection, Base64, ROT13, Leetspeak
-from deepteam.attacks.multi_turn import LinearJailbreaking, CrescendoJailbreaking
+from somewhere import your_model_callback
-attacks = [
- PromptInjection(), # Direct injection attempts
- Base64(), # Encoded payloads
- Leetspeak(), # Character substitution
- LinearJailbreaking(), # Multi-turn escalation
-]
+owasp = OWASPTop10(categories=["LLM_01"])
+attacks = owasp.attacks
+vulnerabilities = owasp.vulnerabilities
-risk_assessment = red_team(
- model_callback=your_model_callback,
- attacks=attacks
+# Modify attributes for your specific testing context if needed
+red_team(
+ model_callback=your_model_callback,
+ attacks=attacks,
+ vulnerabilities=vulnerabilities
)
```
@@ -95,19 +94,20 @@ Prompt injection attacks in DeepTeam test your LLM's resilience against various
- **Intellectual Property**: Proprietary algorithms and trade secrets
- **Authentication Data**: API keys, passwords, tokens
-```python title="main.py"
-from deepteam.vulnerabilities import PIILeakage, PromptLeakage, IntellectualProperty
+```python
+from deepteam.frameworks import OWASPTop10
+from deepteam import red_team
+from somewhere import your_model_callback
-sensitive_vulnerabilities = [
- PIILeakage(types=["direct disclosure", "session leak"]),
- PromptLeakage(types=["secrets and credentials", "instructions"]),
- IntellectualProperty(types=["patent disclosure", "copyright violations"])
-]
+owasp = OWASPTop10(categories=["LLM_02"])
+attacks = owasp.attacks
+vulnerabilities = owasp.vulnerabilities
-risk_assessment = red_team(
- model_callback=your_model_callback,
- vulnerabilities=sensitive_vulnerabilities,
- attacks=[PromptInjection(), Roleplay()]
+# Modify attributes for your specific testing context if needed
+red_team(
+ model_callback=your_model_callback,
+ attacks=attacks,
+ vulnerabilities=vulnerabilities
)
```
@@ -126,21 +126,20 @@ Testing for sensitive information disclosure focuses on specific vulnerability t
- **Library Dependencies**: Vulnerable Python packages or ML frameworks
- **Plugin Ecosystem**: Malicious or compromised LLM plugins
-```python title="main.py"
-from deepteam.vulnerabilities import Bias, Toxicity, Misinformation, Robustness
+```python
+from deepteam.frameworks import OWASPTop10
+from deepteam import red_team
+from somewhere import your_model_callback
-# Test for signs of compromised supply chain components
-supply_chain_tests = [
- Bias(types=["race", "gender", "politics"]),
- Toxicity(types=["profanity", "insults"]),
- Misinformation(types=["factual errors"]),
- Robustness(types=["hijacking"])
-]
+owasp = OWASPTop10(categories=["LLM_03"])
+attacks = owasp.attacks
+vulnerabilities = owasp.vulnerabilities
-risk_assessment = red_team(
- model_callback=your_model_callback,
- vulnerabilities=supply_chain_tests,
- attacks=[PromptInjection(), Roleplay(), LinearJailbreaking()]
+# Modify attributes for your specific testing context if needed
+red_team(
+ model_callback=your_model_callback,
+ attacks=attacks,
+ vulnerabilities=vulnerabilities
)
```
@@ -159,22 +158,20 @@ DeepTeam detects the behavioral impact of supply chain vulnerabilities. If your
- **RAG Poisoning**: Malicious documents in retrieval knowledge bases
- **Embedding Poisoning**: Corrupted vector representations
-```python title="main.py"
-from deepteam.vulnerabilities import (
- Bias, Toxicity, Misinformation, IllegalActivity, GraphicContent, PersonalSafety
-)
+```python
+from deepteam.frameworks import OWASPTop10
+from deepteam import red_team
+from somewhere import your_model_callback
-poisoning_vulnerabilities = [
- Bias(types=["race", "gender", "religion"]),
- Toxicity(types=["profanity", "threats"]),
- Misinformation(types=["factual errors", "unsupported claims"]),
- IllegalActivity(types=["cybercrime", "violent crimes"])
-]
+owasp = OWASPTop10(categories=["LLM_04"])
+attacks = owasp.attacks
+vulnerabilities = owasp.vulnerabilities
-risk_assessment = red_team(
- model_callback=your_model_callback,
- vulnerabilities=poisoning_vulnerabilities,
- attacks=[PromptInjection(), Roleplay(), CrescendoJailbreaking()]
+# Modify attributes for your specific testing context if needed
+red_team(
+ model_callback=your_model_callback,
+ attacks=attacks,
+ vulnerabilities=vulnerabilities
)
```
@@ -189,18 +186,20 @@ risk_assessment = red_team(
- **SQL Injection**: Database queries constructed from LLM output
- **Command Injection**: System commands generated by LLM
-```python title="main.py"
-from deepteam.vulnerabilities import ShellInjection, SQLInjection
+```python
+from deepteam.frameworks import OWASPTop10
+from deepteam import red_team
+from somewhere import your_model_callback
-dangerous_output_tests = [
- ShellInjection(types=["command_injection", "system_command_execution"]),
- SQLInjection(types=["blind_sql_injection", "union_based_injection"])
-]
+owasp = OWASPTop10(categories=["LLM_05"])
+attacks = owasp.attacks
+vulnerabilities = owasp.vulnerabilities
-risk_assessment = red_team(
- model_callback=your_model_callback,
- vulnerabilities=dangerous_output_tests,
- attacks=[PromptInjection(), GrayBox()]
+# Modify attributes for your specific testing context if needed
+red_team(
+ model_callback=your_model_callback,
+ attacks=attacks,
+ vulnerabilities=vulnerabilities
)
```
@@ -218,20 +217,20 @@ Implement output validation in your model callback to catch and sanitize dangero
- **Permissions**: LLM operates with elevated privileges
- **Autonomy**: LLM makes decisions without appropriate oversight
-```python title="main.py"
-from deepteam.vulnerabilities import ExcessiveAgency, RBAC, BFLA, BOLA
+```python
+from deepteam.frameworks import OWASPTop10
+from deepteam import red_team
+from somewhere import your_model_callback
-agency_tests = [
- ExcessiveAgency(types=["functionality", "permissions", "autonomy"]),
- RBAC(types=["role bypass", "privilege escalation"]),
- BFLA(types=["function_bypass", "authorization_bypass"]),
- BOLA(types=["object_access_bypass", "cross_customer_access"])
-]
+owasp = OWASPTop10(categories=["LLM_06"])
+attacks = owasp.attacks
+vulnerabilities = owasp.vulnerabilities
-risk_assessment = red_team(
- model_callback=your_model_callback,
- vulnerabilities=agency_tests,
- attacks=[PromptInjection(), Roleplay()]
+# Modify attributes for your specific testing context if needed
+red_team(
+ model_callback=your_model_callback,
+ attacks=attacks,
+ vulnerabilities=vulnerabilities
)
```
@@ -246,22 +245,20 @@ risk_assessment = red_team(
- **Guards**: Security mechanisms and content filtering rules
- **Permissions and Roles**: Access control configurations
-```python title="main.py"
-from deepteam.vulnerabilities import PromptLeakage
+```python
+from deepteam.frameworks import OWASPTop10
+from deepteam import red_team
+from somewhere import your_model_callback
-prompt_tests = [
- PromptLeakage(types=[
- "secrets and credentials",
- "instructions",
- "guards",
- "permissions and roles"
- ])
-]
+owasp = OWASPTop10(categories=["LLM_07"])
+attacks = owasp.attacks
+vulnerabilities = owasp.vulnerabilities
-risk_assessment = red_team(
- model_callback=your_model_callback,
- vulnerabilities=prompt_tests,
- attacks=[PromptInjection(), PromptProbing(), Base64()]
+# Modify attributes for your specific testing context if needed
+red_team(
+ model_callback=your_model_callback,
+ attacks=attacks,
+ vulnerabilities=vulnerabilities
)
```
@@ -280,19 +277,20 @@ Never include sensitive credentials or secrets directly in system prompts. Use e
- **Vector Database Access**: Unauthorized access to embedding stores
- **Embedding Inversion**: Reconstructing source text from vectors
-```python title="main.py"
-from deepteam.vulnerabilities import Misinformation, PIILeakage
+```python
+from deepteam.frameworks import OWASPTop10
+from deepteam import red_team
+from somewhere import your_model_callback
-# Test for RAG-specific vulnerabilities
-rag_tests = [
- Misinformation(types=["factual errors"]), # May indicate poisoned knowledge base
- PIILeakage(types=["direct disclosure"]) # May indicate vector database leakage
-]
+owasp = OWASPTop10(categories=["LLM_08"])
+attacks = owasp.attacks
+vulnerabilities = owasp.vulnerabilities
-risk_assessment = red_team(
- model_callback=your_model_callback,
- vulnerabilities=rag_tests,
- attacks=[PromptInjection(), GrayBox()]
+# Modify attributes for your specific testing context if needed
+red_team(
+ model_callback=your_model_callback,
+ attacks=attacks,
+ vulnerabilities=vulnerabilities
)
```
@@ -311,22 +309,20 @@ Vector and embedding security is crucial for RAG systems. Ensure proper access c
- **Expertise Misrepresentation**: False claims about qualifications
- **Fabricated Sources**: Made-up citations, studies, or references
-```python title="main.py"
-from deepteam.vulnerabilities import Misinformation, Competition
+```python
+from deepteam.frameworks import OWASPTop10
+from deepteam import red_team
+from somewhere import your_model_callback
-misinformation_tests = [
- Misinformation(types=[
- "factual errors",
- "unsupported claims",
- "expertize misrepresentation"
- ]),
- Competition(types=["discreditation"])
-]
+owasp = OWASPTop10(categories=["LLM_09"])
+attacks = owasp.attacks
+vulnerabilities = owasp.vulnerabilities
-risk_assessment = red_team(
- model_callback=your_model_callback,
- vulnerabilities=misinformation_tests,
- attacks=[PromptInjection(), Roleplay(), PromptProbing()]
+# Modify attributes for your specific testing context if needed
+red_team(
+ model_callback=your_model_callback,
+ attacks=attacks,
+ vulnerabilities=vulnerabilities
)
```
@@ -345,17 +341,20 @@ LLM misinformation can have serious consequences in domains like healthcare, fin
- **API Abuse**: High-volume requests leading to cost escalation
- **Context Flooding**: Extremely long inputs that overflow context windows
-```python title="main.py"
-# Test for potentially resource-intensive patterns
-resource_test = red_team(
- model_callback=your_model_callback,
- attacks=[
- MathProblem(), # Complex computational requests
- LinearJailbreaking(), # Extended conversations
- CrescendoJailbreaking() # Escalating complexity
- ],
- attacks_per_vulnerability_type=3, # Limited testing
- max_concurrent=1 # Sequential testing to monitor resources
+```python
+from deepteam.frameworks import OWASPTop10
+from deepteam import red_team
+from somewhere import your_model_callback
+
+owasp = OWASPTop10(categories=["LLM_10"])
+attacks = owasp.attacks
+vulnerabilities = owasp.vulnerabilities
+
+# Modify attributes for your specific testing context if needed
+red_team(
+ model_callback=your_model_callback,
+ attacks=attacks,
+ vulnerabilities=vulnerabilities
)
```