The study focuses on the intersection of cybersecurity and AI reproducibility. It introduces a new framework for evaluating AI platforms’ reproducibility from a cybersecurity perspective. Five popular AI platforms were assessed, revealing that none fully incorporate the necessary cybersecurity measures for robust reproducibility. The study emphasizes the importance of integrating specific cybersecurity features into AI platforms to address the challenges associated with AI reproducibility. The proposed framework can be applied beyond AI platforms, serving as a tool for evaluating a wide range of systems from a cybersecurity perspective.
Publication date: 28 Sep 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2309.15525
