Access control in DBMSs is critical for ensuring data security and integrity. However, the increasing complexity of its implementation often introduces broken access control (BAC) vulnerabilities. These vulnerabilities can lead to severe consequences, including privilege escalation, unauthorized data access, or even full compromise of the DBMS. Existing manual testing for BAC vulnerabilities is time-consuming and incomplete. Automated methods like static analysis also struggle in DBMSs, as static rules are difficult to apply to multi-level and dynamically changing privileges. In this paper, we propose Beacon , which detects BAC vulnerabilities by validating the consistency between SQL operations and system catalogs. Our key insight is that the visibility of objects in the system catalogs is consistent with the user’s access control: if an object is invisible to a user in the system catalogs, the user should not have any access privileges on that. Any inconsistency suggests that a user is exceeding their privileges, indicating a potential BAC vulnerability.We used Beacon to test eight popular DBMSs (e.g., MySQL and MariaDB), uncovering 39 previously unknown BAC vulnerabilities. Among them, 19 result in privilege escalation, and 20 lead to unauthorized information disclosure. Moreover, 7 of them have existed in DBMSs for more than 6 years, with the longest-persisting one lasting 13 years. DBMS vendors took these issues seriously and have already confirmed all of these vulnerabilities. Many vendors provided positive feedback, recognizing the importance of addressing these vulnerabilities. For instance, OceanBase awarded bounties for reported vulnerabilities, underscoring Beacon ’s role in improving DBMS access control.

@article{paperbot3790,
  title = {Beacon: Detecting Broken Access Control Vulnerabilities in DBMSs via System Catalog Consistency Validation},
  author = {Zongrui Peng and Jingzhou Fu and Zhiyong Wu and Jie Liang and Xiangdong Huang and Dalong Shi and Yu Jiang},
  journal = {Proceedings of the ACM on Programming Languages},
  volume = {10},
  number = {OOPSLA1},
  year = {2026},
  doi = {10.1145/3798232}
}