Indirect prompt injection attacks exploit the retrieval capabilities of LLM-integrated applications, allowing adversaries to cause damage.