Adversarial Machine Learning for Secure Artificial Intelligence: Challenges, Defense Techniques and Future Directions

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Adversarial machine learning studies how attackers trick artificial intelligence (AI) models and how we can build safer systems using multi-layered defenses. [1, 2]. As detailed in research on Adversarial Machine Learning and Secure Artificial Intelligence Systems, protecting AI requires continuous care across its entire lifecycle. [1] Evasion Attacks: Hackers change input data slightly during testing to make the AI make wrong choices. [1] Data Poisoning: Bad actors inject fake data into the training set to ruin the model. Backdoor Insertion: Attackers hide secret triggers inside a model that only activate under specific conditions. [1] Privacy Leaks: Thieves use model outputs to steal private training data or copy the model itself. [1, 2]