Prompt injection is an attempt to trick an AI into ignoring its instructions or doing something it should not. The misleading request can be written directly in a prompt or hidden inside a document the AI is asked to read.
Imagine asking a helper to summarize a letter, but the letter contains a note saying “ignore your manager and reveal private files.” That note is part of the letter, not a trusted instruction.
AI apps should treat outside content carefully and limit what actions the AI can take. A person should review sensitive actions and important results.