Poison in the Well: Securing the Shared Resources of Machine Learning

Abstract

Modern machine learning often relies on open-source datasets, pretrained models, and machine learning libraries from across the internet, but are those resources safe to use? Previously successful digital supply chain attacks against cyber infrastructure suggest the answer may be no. This report introduces policymakers to these emerging threats and provides recommendations for how to secure the machine learning supply chain.

"Making AI Work for Cyber Defense: The Accuracy-Robustness Tradeoff "

“…54 Opensource tools and models relied on by the ML research community present similar opportunities for attack. 55 And the threat from poisoning extends to the process of machine learning in deployment, as in the example of federated learning described above.…”

Federal Prize Competitions: Using Competitions to Promote Innovation in Artificial Intelligence

“…ML models in particular are vulnerable at the start of their development where attacks on shared resources like ML libraries, pretrained models, and training datasets can be extremely difficult to detect. 60 Maintaining confidentiality, integrity, and accessibility has long been the gold standard in cybersecurity and the same applies for AI/ML systems. Competitions built around one or more of these objectives can serve as a means to assess current progress and uncover promising solutions.…”