Akademik Veri Yönetim Sistemi
2025 Innovations in Intelligent Systems and Applications Conference, ASYU 2025, Bursa, Türkiye, 10 - 12 Eylül 2025, (Tam Metin Bildiri)
The introduction of large language models into our lives has created a revolution, allowing artificial intelligence to be easily used and accessible by everyone. With this opportunity, large language models are now used in every step of our lives, from education to economy, agriculture to health, and are on their way to becoming a part of us. This widespread use of large language models attracts the attention of malicious people and brings with it security problems and concerns. Despite this rapid growth and expanding attack surface area, the precautions taken by model publishers are inadequate. This study focuses on the current inadequacy in the relevant field and presents a detection suggestion against 'Prompt Injection' attacks, one of the most commonly used attacks against large language models. Within this scope, 10 different models developed using various approaches were used to analyze the input text and classify whether or not it contained attacks. The tests showed that the most successful results were obtained with the BERT-based model, achieving 88 % accuracy and an F1 score of 87 %. The most important feature that distinguishes the study from existing studies is that, unlike existing detection models, it has the ability to detect in Turkish.