Turkish Email Phishing Detection Using Fine Tuned Large Language Models

Phishing emails are one of the most common types of attack used by adversaries worldwide. The adversaries prepare targeted phishing emails for specific users who generally use their native languages. On the other hand, phishing email detection mechanisms designed by global vendors are generally capable to detect phishing emails written in English. This circumstance is a big challenge for organizations that require detecting phishing emails in native languages, such as in Turkish. In this research, we propose a novel fine-tuned Large Language Model to detect Turkish phishing emails. We experimentally evaluated the proposed model with real phishing emails in Turkish. Analysis results show that the proposed model detects phishing emails in Turkish with high accuracy, which is not observed with global vendors phishing detection mechanisms for Turkish emails. Thus, native language based phishing detection mechanisms may provide better detections for phishing emails for native languages, such as Turkish.