Improper Neutralization of Input Used for LLM Prompting (CWE-1427)
LLM Prompt Injection & Jailbreaking Cheat Sheet
Prompt Injection is a vulnerability where an attacker manipulates an LLM's output by crafting malicious inputs that override its original system instructions
💉 Prompt Injection & Jailbreaking Cheat Sheet
Prompt Injection is a technique used to manipulate the output of Large Language Models (LLMs) by crafting inputs that override or bypass the model's original system instructions. This cheat sheet provides a comprehensive overview of vectors, payloads, and real-world examples for educational and defensive purposes.
📖 1. Direct Prompt Injection
Direct Prompt Injection occurs when an attacker inputs malicious instructions directly into the LLM prompt, attempting to hijack the model's behavior, leak the system prompt, or expose internal configurations.