**Abstract:** A significant milestone has been achieved by Wuhan University's National Cybersecurity College, where a research paper co-authored by a 2021 PhD student, Ge Yunjie, and a 2020 undergraduate student, Chen Pinji, from the Cybersecurity Experimental Class has been accepted for presentation at the 33rd USENIX Security Symposium in 2024. This marks the first time an undergraduate student from Wuhan University has been a first or co-first author on a paper published in one of the top four international cybersecurity conferences—USENIX Security, IEEE S&P, ACM CCS, and NDSS. The paper, titled "More Simplicity for Trainers, More Opportunity for Attackers: Black-Box Attacks on Speaker Recognition Systems by Inferring Feature Extractor," was supervised by Professor Wang Qian and Associate Professor Zhao Lingchen, with contributions from Professor Wang Cong of City University of Hong Kong, Associate Professor Li Qi of Tsinghua University, and Professor Shen Chao of Xi'an Jiaotong University. Other contributors include PhD students Mu Ningping and Jiang Peipei from the National Cybersecurity College. The research focuses on the vulnerabilities of speaker recognition systems (SRS) used in various critical applications such as identity verification, telephone banking, and financial transactions. These systems are susceptible to black-box adversarial attacks, which are often limited by low success rates and high query costs. The authors identified a design flaw in the feature extractor module of SRS, noting that certain algorithms are frequently reused across different systems. Leveraging this insight, they developed a novel feature extractor inference technique using a genetic algorithm, which can successfully replicate the target system's feature extractor with approximately 300 queries. This method significantly enhances the accuracy and efficiency of black-box attacks. Building on this foundation, the authors introduced a technique to generate adversarial samples in the feature space, integrating psychoacoustic models and room impulse response (RIR) technology. These integrations not only improve the attack's success rate but also enhance its stealth and robustness in real-world physical environments, thereby reducing the overall deployment cost. The effectiveness of their approach was demonstrated through successful attacks on four major cloud-based speaker recognition services and seven commercial voice-controlled devices. The findings of this research highlight critical design flaws in current speaker recognition systems, posing new challenges to their security and architecture. The paper underscores the importance of robust security design in the development of SRS and other AI-driven authentication mechanisms. The acceptance of this paper by USENIX Security, a conference with a historically low acceptance rate of 19% over the past decade, underscores the high quality and significance of the research conducted by Wuhan University's team. USENIX Security, established in 1990, is one of the most prestigious and long-standing conferences in the field of cybersecurity. It is recognized by the Chinese Computer Federation (CCF) as an A-class conference, reflecting the cutting-edge research and contributions it consistently showcases. The involvement of undergraduate students in such high-level research and publication is a testament to the strong educational and research environment at Wuhan University, particularly within its National Cybersecurity College. This achievement not only highlights the university's commitment to fostering young talent in cybersecurity but also positions Wuhan University as a leading institution in the field. The research contributes to the broader understanding of the security risks associated with AI-based authentication systems and provides valuable insights for the development of more secure and resilient technologies.