AI Supply Chain Attacks

#GenerativeAI #bills-of-materials #defense #integrity

Join us for the AI Supply Chain Attacks session to explore how vulnerabilities across the AI lifecycle can compromise models, datasets, development environments, CI/CD pipelines, and deployed AI systems. The session will examine how attackers exploit vulnerable third-party packages, poisoned datasets, compromised pre-trained models, transfer-learning processes, public model repositories, and insecure build and deployment pipelines. Participants will learn about common attack techniques, including data and model poisoning, hidden backdoors, dependency confusion, model tampering, and malicious code embedded within model files.

The session will include a case study demonstrating how a compromised AI component can move through a CI/CD pipeline and reach production. It will also introduce practical defense strategies, including vulnerability and model scanning, trusted private repositories, integrity verification, model and data provenance, governance and lineage, AI/ML Bills of Materials, continuous monitoring, and secure MLOps workflows.

Session outline