3

We Don't Need a Large Language Model for That: When Traditional Machine Learning and Encoders Beat Decoder LLMs in Security Classification