← All articles

Database vulnerabilities let low-privilege accounts run code

Database vulnerabilities let low-privilege accounts run code

Database vulnerabilities in four database systems can allow code to run on the database server, even when an account cannot read application data. A researcher at Depthfirst demonstrated proof-of-concept exploits against three of the systems, plus a separate attack chain against the fourth that needs no login. None of the five issues has been observed in a real attack.

The research concerns memory corruption, a programming error that can make software write or use data in the wrong part of memory. The affected database engines use memory-unsafe C and C++ code to read, plan and execute ordinary SQL, the language used to work with database information. The researcher’s point is that SQL permissions mainly control access to tables and schemas, not necessarily the engine code reached while processing a query.

Common permissions can provide that route. One system’s default temporary-table privilege, along with permission to create temporary tables in two others, may allow an account to create objects or submit input that reaches a memory bug. All five flaws now have vendor fixes, four have CVE identifiers, and the complete attack chains have not been independently reproduced.

In the first database system, CVE-2026-116331 allows an account that can create database objects to execute code as the operating-system user running the database. The issue was fixed in a security advisory on November 12. The researcher’s analysis describes a normal login role creating the required objects in its own temporary schema, without accessing application data, then using a type-confusion bug, where software treats one kind of data as another, to gain control of the server.

The second database system has a related flaw in its query optimizer, the part of the database that chooses how to run a request. CVE-2026-101786 affects database server and cluster versions 26.7.0 through 26.7.1 and was fixed on October 20. The vendor’s advisory rates the issue as requiring high privileges. However, the researcher reached code execution on a 26.7.0 image using an account that could create temporary tables but could not read existing tables, a narrower level of access than that rating might suggest.

The commercial database presents a different risk. According to the researcher, an attacker does not need an account when two weaknesses are chained against an exposed web endpoint. A bug in its text-search component and a flaw in the GraphQL interface of its REST data services can reach database engine code and run commands before authentication. The vendor patched CVE-2026-114819 in the text-search component on November 17. It carries a severity rating of 9.9 and affects database versions 19.3 through 19.32, 21.3 through 21.24 and 23.4.0 through 23.26.3. The GraphQL issue, CVE-2026-91609, affects the REST data services 24.2.0 through 26.2.1 and was fixed on October 20. The vendor rates neither issue, on its own, as exploitable without a login.

The fourth database is the other target. The researcher reported an integer-overflow write, where a calculation exceeds the space available for its result, in code that reads a binary row format. A user can reach it by setting the format_binary_max_string_size option to zero. The researcher reproduced a shell, meaning a command-running interface, on version 26.7.5.10.

The project told The Hacker News that the issue was reported through its bug bounty programme and fixed in early September. It has no CVE assigned. The open-source fix is included in version 26.9 and later, and was backported to older release lines in builds 26.8.3.41, 26.7.7.30, 26.6.5.55 and 26.3.32.8. An earlier report that the issue had no confirmed fixed version was corrected on November 20, 2026.

Administrators should install the relevant updates and review what database accounts can do, not only which data they can read. Removing the temporary-table privilege and broad temporary-table permissions from accounts that do not need them can reduce exposure to the first, second and fourth paths. That measure does not address the commercial database chain, which begins without an account, so exposed web endpoints also require attention.

For organisations that need broader help reviewing infrastructure and access risks, AEU-I provides security-first IT, infrastructure and consulting. The immediate priority for database operators remains applying the vendor fixes and limiting unnecessary permissions.

Also available in: DeutschItalianoShqip