AI vs. AI: The Evolution of Offensive and Defensive AI Techniques in Cybersecurity
Dr. Felix Hernandez · 2025
The research investigates the development of AI systems and their artificial counterparts for cybersecurity purposes by studying defensive and offensive techniques that strengthen security and execute automated cyberattacks. This study evaluates four machine learning models, including Random Forest along with Support Vector Machines (SVM) and Autoencoders and Adversarially Trained Models using five performance metrics such as accuracy and precision, recall, and F1 score as well as Adversarial Resilience Score (ARS). The adversarial model demonstrates superior results, which yield 94% accuracy together with 0.92 precision, 0.93 recall, and a 0.92 F1 score. Random Forest demonstrated 92% percentage accuracy and reached a precision value of 0.90, while recall reached 0.91, and the F1 score equaled 0.90. The accuracy rate for SVM reached 89%. The autoencoder demonstrated the worst performance as it resulted in 87% precision together with 0.85 precision and 0.86 recall. The adversarial model showcased high reliability through its ARS value of 0.91.