Prompt Injection vulnerabilities occur in AI/LLM-powered applications where user input can manipulate the system's prompts or instructions. This can lead to unauthorized actions, data leakage, or bypassing of security controls.
- System prompt extraction
- Instruction override
- Jailbreaking AI models
- Context manipulation
- Role confusion attacks
- Indirect prompt injection via external data
Test AI-powered chatbots, assistants, and applications that use Large Language Models (LLMs). Try to manipulate the model's behavior by injecting malicious prompts that override system instructions.
See prompt-injection-payloads.txt for a comprehensive list of prompt injection payloads.