What are the definitions for prompt injections, jailbreaks, prompt hacking and prompt leaking?

Below are concise meanings of several common terms you'll run into when working on advanced prompts:

• Prompt injection – Inserting crafted text into the input or context to override or influence the model's original instructions.
• Jailbreak – A specialised prompt injection designed to bypass content filters or system policies so the model produces otherwise restricted output.
• Prompt hacking – The broader practice of analysing, manipulating, or optimising prompts (including benign optimisation and adversarial attacks).
• Prompt leaking – Forcing the model to reveal hidden developer or system prompts that were not meant to be disclosed.