A Comprehensive Study: How Attackers Can Exploit Vulnerabilities in Al Models and Develop Robust Cyber Strategies with Experimental Validation on CIFAR-10

Read full text Summarise with VIRAT
Cited by: 0
References: 10

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Artificial Intelligence (AI) plays a crucial role in modern cybersecurity.It allows for quick detection of unusual activities, prevention of intrusions, and forecasting of potential threats.However, AI models are increasingly becoming targets for skilled attackers who take advantage of their weaknesses.These attackers use methods like adversarial examples, data poisoning, model theft, and privacy attacks to compromise the integrity, confidentiality, and availability of AI systems.This study examines how adversaries exploit weaknesses in AI models and suggests strong strategies to combat these threats.Along with a review of adversarial machine learning (AML) techniques, we provide an assessment of adversarial attacks and defenses on a ResNet-18 model trained using the CIFAR-10 dataset.By applying the Fast Gradient Sign Method (FGSM), we show that even a well-performing ResNet-18 model is vulnerable to adversarial changes, resulting in an accuracy drop from 85.86% to 14.20% at ϵ=0.05.We also demonstrate that adversarial training boosts the model's resilience, as the protected model maintains much higher accuracy under attack while achieving 81.53% accuracy on clean data.Our results indicate that securing AI requires both technical measures and governance systems.This study combines a classification of threats and case studies with experimental validation to give a complete overview of adversarial risks and defenses in AI-focused cybersecurity.

Fields of Study & Subjects

Adversarial Robustness in Machine Learning Security and Verification in Computing Digital Media Forensic Detection

Publication details

Journal / source
International Research Journal of Modernization in Engineering Technology and Science
Publication date
31 March 2026
Language
EN