Insane in the Supply Chain: Threat modeling for...
Eoin Wickens, Marta Janus
BSidesSF 2024 · Day 1
This talk, "Insane in the Supply Chain: Threat modeling for attacks on AI systems," delivered by Eoin Wickens and Marta Janus, researchers at Hidden Layer's Synaptic Adversarial Intelligence (SAI) team, delves into the critical and rapidly evolving landscape of security threats targeting the artificial intelligence supply chain. Building upon their previous work introducing attacks on AI and machine learning, this presentation focuses on the specific vulnerabilities and attack vectors that emerge when AI systems are integrated into broader software development and deployment pipelines. The speakers highlight the urgent need for robust threat modeling in an era where AI, particularly Large Language Models (LLMs), is being rapidly adopted across virtually every product and service imaginable.
AI review
This talk provides a solid, technically grounded overview of the AI supply chain's most critical attack vectors. The speakers dissect various vulnerabilities, from data poisoning to insecure model serialization and MLOps tooling exploits, with concrete examples and their own research. While some concepts aren't entirely novel, the comprehensive threat modeling and specific, recent findings make this a valuable session for anyone serious about securing AI systems.