From MLOps to MLOops - Exposing the Attack Surface of Machine Learning Platforms

Explore a comprehensive Black Hat conference talk that delves into the security vulnerabilities of MLOps platforms in organizations. Learn how popular open-source ML platforms like MLflow, Kubeflow, and Metaflow, while streamlining AI and ML processes, can become potential security risks. Discover detailed analyses of six major OSS MLOps platforms, examining how their features can be exploited for organizational attacks. Understand server-side and client-side CVEs affecting both platform servers and clients, including data scientists and MLOps CI/CD machines. Gain critical insights into inherent vulnerabilities within MLOps platform formats that pose risks even in fully patched systems. Master essential knowledge for both red and blue teams about secure MLOps platform deployment, proper user briefing, and potential attack vectors for each platform feature. Presented by Shachar Menashe, Sr. Director Security Research at JFrog, this 38-minute security analysis provides valuable information for organizations implementing AI and ML technologies.