prompt injection

A **prompt injection** is a type of attack against an AI language model (like ChatGPT) where malicious or deceptive instructions are inserted into the input text (the “prompt”) to manipulate the model’s behavior. The goal is to make the model ignore its original instructions, reveal confidential information, or perform unintended actions. Prompt injections can appear in user inputs, embedded text, or data sources, and defending against them involves filtering inputs, constraining model capabilities, and validating outputs.