Pipelines of Privilege: Attack Paths from DevOps to MLOps Infrastructure | SO-CON 26

BLUF: This presentation explores attack paths from DevOps to MLOps, highlighting how traditional infrastructure vulnerabilities can compromise machine learning models, and outlines defensive strategies to secure the ML lifecycle [0:00].

Key Takeaways:
• The speaker, Brett Hawkins from Armory, introduces the topic of attack vectors bridging DevOps and MLOps, noting the session will be demo-heavy to illustrate these risks [0:00].
• A significant portion of the talk is dedicated to demonstrating how attackers can exploit the ML pipeline, with sections three and four focusing on the bulk of these offensive demonstrations [0:19].
• The presentation shifts to defensive measures in sections five and six, aiming to provide a comprehensive wrap-up on securing ML systems against these identified attack paths [0:23].
• The speaker references a recently released blog post detailing this research, encouraging the audience to review additional written material for deeper context [0:27].

Closing Statement: Understanding the intersection of DevOps and MLOps security is critical for protecting machine learning infrastructure from sophisticated attack paths.

Sources:

• 0:00 Introduction to the topic of DevOps to MLOps attack paths.
• 0:19 Overview of the demo-heavy structure focusing on offensive techniques.
• 0:23 Transition to defensive strategies and conclusion.
• 0:27 Reference to a related blog post for further reading.