Indirect Prompt Injection

A security attack method in which an attacker hides malicious instructions in external content that a model reads, such as web pages, code comments or tool return results, to induce the model to perform unintended actions.