With the rapid advancement and widespread adoption of AI, cloud computing has gained renewed momentum. At the foundation of secure cloud infrastructure lies a bug-free hypervisor. Embracing the philosophy of full-lifecycle security enforcement, this talk presents our research on identifying and fixing hypervisor vulnerabilities before release, along with complementary efforts to mitigate attacks in production environments. This talk not only summarizes current technical approaches but also highlights emerging research directions aimed at achieving broader AI system security.

Qiang Liu is a postdoc at EPFL, working with Prof. Mathias Payer in the HexHive laboratory. He earned his Ph.D. in 2023 from Zhejiang University (ZJU) under the guidance of Prof. Yajin Zhou. His research in cybersecurity focuses on 1) developing prior-to/after-release security enforcement of software based on our deep understanding, and 2) building the chain of trust examined by full-chain exploits. During his PhD and PostDoc, Qiang Liu pioneered innovative solutions, including auto-extracted input dependencies and a generic executor, to significantly strengthen hypervisor security. These breakthroughs led to the discovery and mitigation of around 100 hypervisor vulnerabilities prior to software release. His work has been recognized at all the top security conferences: IEEE S&P, Usenix Security, ACM CCS, and ISOC NDSS. He received the Best Paper Awards at USENIX Security'24 and ACM RAID'24. He is also serving on the program committee for IEEE/ACM ASE'25 and USENIX Security'25 and is a reviewer for ACM CSUR and ACM TOSEM.