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.mdxindex 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 OWASPTop10from 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)```