graph TD
A[End Users] --> B[Application Services]
B --> C[LLM Application]
C --> D[LLM Production Services]
D --> E[LLM Automation Agents]
D --> F[LLM Modes]
F --> G[Training Dataset and Processing]
G --> H[Fine-Tuning Data]
G --> I[Training Data]
I --> J[External Data Sources]
H --> J
D --> K[Plugins or Extensions]
K --> L[Downstream Services]
L --> M[Database]
L --> N[Websites]
L --> O[Services]
- Prompt Injection
- Jailbreak e.g DAN (Do/Say Anything Now, )
-
Defense – Guardrails / AI Firewalls / – Garak scanner for AI
- Red Team Strategy - Johann’s Blog
- OWASP Top 10 for LLM Applications
- Getting Started with AI CTFs
- Real-world Attacks on LLM Applications
- ChatGPT for Red Teams
- Attacking and Defending Generative AI
- AI Vulnerability Database