Adversarial Attacks and Defenses in AI Systems: Challenges, Strategies, and Future Directions

AI systems are vulnerable to adversarial manipulations (Szegedy et al., 2014). These attacks exploit model weaknesses through subtle input perturbations (Carlini & Wagner, 2017), risking safety in applications like facial recognition and autonomous driving (Eykholt et al., 2018). Defense mechanisms, including adversarial training (Madry et al., 2018) and input preprocessing (Guo et al., 2018), often face trade-offs between robustness and efficiency.