Potential Applications of GPT-4o for Cyber Threat Analysis and Detection

This bachelor’s thesis comprehensively explores the use of the GPT-4 model for cyber threat analysis and detection. The theoretical part establishes the foundational concepts of cybersecurity, the architecture and historical evolution of large language models, and their applications in threat detection. A detailed literature review summarizes existing studies on malware analysis, phishing, DDoS attacks, and SQL injection. The practical section presents a series of experiments evaluating GPT-4o’s performance on real and synthetic datasets in zero-shot and few-shot scenarios. Malware, phishing, DDoS, and SQL injection were assessed using accuracy, precision, recall, and F1-score metrics. Results demonstrate that GPT-4o can to a certain extent identify diverse attack patterns and that few-shot prompting significantly improves detection success. The findings highlight the promise of these models in augmenting traditional defense tools while underscoring limitations related to data quality and the absence of fine-tuning.