Adversarial Robustness in XAI Specifications: Why Explainability Must Be Secure

Key Points

• To examine the relationship between adversarial robustness and the security of explainable AI specifications.
• Identified vulnerabilities in existing explainable AI frameworks
• Evaluated the robustness of various machine learning models under adversarial attacks
• Assessed security measures in explainability specifications.
• Found that many explainable AI models exhibit significant vulnerabilities under adversarial conditions.
• Showed that integrating security measures enhances the robustness of explainable AI systems.

Abstract