An attack where malicious instructions are inserted into user input to manipulate an AI model's behavior.