Cheatsheet Summary
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 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.
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.
No comments yet.
Be the first to share your thoughts
Log in to join the discussion.
Sign In