Baojun Liu

Modern cellular networks rely on DNS-based service discovery to select internal control-plane and data-plane functions. Such queries are intended to remain within operators' private namespaces, yet misconfigurations can leak them to the public Internet. Despite warnings from 3GPP and IETF decades ago, the prevalence and security implications of such leakage remain unexamined. We present the first large-scale measurement of cellular service-discovery DNS leakage. By analyzing two days of B-root traffic for each year from 2021 to 2025 and identifying internal cellular queries via 3GPP naming patterns, we observed 13.46 million leaked queries from 139 countries, covering 135,845 unique FQDNs. Most leakage (93.10%) was sporadic, suggesting operators remediated issues over time, yet a small set showed persistent leakage across all five years. Leakage frequently arose in cross-operator and cross-country scenarios, consistent with roaming behavior. While the increased load on B-root is minimal (0.02% on average), leaked names often encode internal deployment parameters (e.g., base-station identifiers), exposing information that operators do not otherwise publish. Using open-source testbeds, we further show that an on-path adversary can redirect service-discovery responses once such queries leave the operator's network, which would otherwise be infeasible without the leak. Our work highlights the need for stricter DNS isolation in cellular networks.

Public Key Infrastructure (PKI) secures Internet communications, and certificate pinning strengthens authentication by binding trusted certificates or keys to clients. However, its deployment is complex, which may lead to communication failures or weakened security guarantees. In this study, we conduct a large-scale analysis of 18,528 Android applications to investigate the deployment and security of certificate pinning. We find that only 115 applications adopt NSC-based pinning, among which 90.4% exhibit misconfigurations. Overall, we highlight the gap between secure design and real-world deployment, and discuss best practices to improve both security and usability in mobile applications.

Web agents, powered by large language models (LLMs), are increasingly deployed to automate complex web interactions. The rise of open-source frameworks (e.g., Browser Use, Skyvern-AI) has accelerated adoption, but also broadened the attack surface. While prior research has focused on model threats such as prompt injection and backdoors, the risks of social engineering remain largely unexplored. We present the first systematic study of social engineering attacks against web automation agents and design a pluggable runtime mitigation solution. On the attack side, we introduce the AgentBait paradigm, which exploits intrinsic weaknesses in agent execution: inducement contexts can distort the agent's reasoning and steer it toward malicious objectives misaligned with the intended task. On the defense side, we propose SUPERVISOR, a lightweight runtime module that enforces environment and intention consistency alignment between webpage context and intended goals to mitigate unsafe operations before execution. Empirical results show that mainstream frameworks are highly vulnerable to AgentBait, with an average attack success rate of 67.5% and peaks above 80% under specific strategies (e.g., trusted identity forgery). Compared with existing lightweight defenses, our module can be seamlessly integrated across different web automation frameworks and reduces attack success rates by up to 78.1% on average while incurring only a 7.7% runtime overhead and preserving usability. This work reveals AgentBait as a critical new threat surface for web agents and establishes a practical, generalizable defense, advancing the security of this rapidly emerging ecosystem. We reported the details of this attack to the framework developers and received acknowledgment before submission.