Organizer
Partner Company
AI Model Fuzzing: Finding Vulnerabilities in TensorFlow
ALT, Kang Li
Baidu Security
Abstract
Due to data scalability and training cost considerations, developers often use third-party AI models to build AI applications. Consequently, the risk of using bad AI models as an attack vector arises. Bad AI models can trigger AI framework vulnerabilities in different phases, including model optimization, training, inference, etc. However, traditional fuzzing methods based on input mutation become ineffective given the complexity of modern AI frameworks. In this talk, we will share our experience and results on fuzzing TensorFlow.
Speaker Info
ALT is a senior security researcher from Baidu Security. He is experienced in both offensive and defensive security on confidential computing, system security, and mobile security. Over the years, he has also discovered various TEE and iOS framework vulnerabilities with 20+ CVEs credited by Google, Microsoft, and Apple.
Kang Li
KANG is the director of Baidu Security Research. He is a frequent speaker at BlackHat and POC. Over the years, he has discovered vulnerabilities in various systems from CDN to DNS, and from Mobile Bootloader to Deep Learning Frameworks.