Defenses

This paper provides a formal framework and benchmarking methodology for prompt injection attacks and their countermeasures in large language models.