Spotlighting is an effective prompt engineering technique to defend against indirect prompt injection attacks on large language models.
