KLUE RoBERTa Based Prompt Injection Detection Model
This repository contains a fine-tuned KLUE RoBERTa based text classification model for Korean prompt injection detection.
Intended Use
This model was developed for a university text mining project. It is intended for academic evaluation and research on Korean prompt injection detection.
The model classifies Korean text into two categories:
normalattack
Evaluation
| Metric | Value |
|---|---|
| Accuracy | 0.9870 |
| Attack Precision | 0.9752 |
| Attack Recall | 0.9882 |
| Attack F1 | 0.9816 |
| Macro F1 | 0.9858 |
| Weighted F1 | 0.9870 |
Detailed evaluation files:
classification_report.txtsummary.csv
Project Repository
https://github.com/3recon/prompt-injection-detection
Limitations
This model is a research artifact for prompt injection detection. It should not be used as a standalone production security system without additional validation.