Researcher shows database code execution with limited access

Researcher shows database code execution with limited access

Add CyberWorldOpsto your preferred sources on Google
Researcher shows database code execution with limited access
Vulnerabilities

Illustrative image generated with AI

Database accounts unable to read application tables can still reach flaws that let them run code on the database server, Depthfirst researcher Zheng Yu demonstrated, The Hacker News reported. The outlet described the exploits as published “yesterday.” The work covers PostgreSQL, MySQL and ClickHouse, plus a separate Oracle attack chain that reportedly requires no login, making table-access restrictions insufficient protection against these paths.

The demonstrations concern organizations running vulnerable database releases, including those that give application accounts permission to create temporary objects. The Hacker News said the full attack chains had not been independently reproduced and that the research reported no observed use in real attacks. Four flaws have vendor-assigned CVE identifiers and fixes; the ClickHouse issue has no assigned CVE or confirmed fixed version.

Why limited SQL permissions can still expose the server

Yu’s central finding is that SQL permissions govern access to tables and schemas, while queries still reach the database engine’s C and C++ code. Memory-corruption bugs—errors in how that code handles memory—can turn permitted database operations into execution of attacker-controlled code.

Common permissions provide a route into those operations. PostgreSQL’s default TEMPORARY privilege, and temporary-table creation rights in MySQL and ClickHouse, can let an account construct objects or submit input that reaches vulnerable engine code without reading application data.

For PostgreSQL, the vendor’s CVE-2026-116331 advisory confirms that an account able to create objects can execute arbitrary code as the operating system user running the database. The flaw affects optimization of minimum and maximum aggregate operations, which calculate a result from multiple input values.

A user-defined aggregate’s SORTOP option makes the query planner incorrectly treat different input and result types as identical. This type confusion—handling a value as the wrong data type—creates the code-execution opportunity. PostgreSQL rates the vulnerability 8.8 on the CVSS severity scale.

The Hacker News described Yu’s demonstration as using an ordinary login role to create the necessary objects inside its own temporary schema, without touching application data.

In MySQL, CVE-2026-101786 affects the query optimizer, the component that chooses how to execute a query. The report identifies MySQL Server and Cluster versions 26.7.0 through 26.7.1 as affected. It says Oracle’s October advisory rates exploitation as requiring high privileges, while Yu demonstrated code execution on a 26.7.0 image using an account permitted only to create temporary tables and unable to read existing tables.

For ClickHouse, Yu reported an integer-overflow write, where a numerical calculation exceeds its permitted range and leads to an unsafe memory write, in the handling of a binary row format. Setting format_binary_max_string_size to zero enabled the path, and the researcher demonstrated a command shell on version 26.7.5.10. That is a tested version, rather than a complete affected-version range.

Oracle chain reportedly bypasses the need for an account

The Oracle demonstration combines an Oracle Text vulnerability with a flaw in the GraphQL interface of Oracle REST Data Services, a web interface to database services. Yu reportedly used an exposed web endpoint to reach database engine code and execute commands before authentication.

The Hacker News identifies the Oracle Text issue as CVE-2026-114819, rated 9.9 and patched November 17, and the GraphQL issue as CVE-2026-91609, patched October 20. It reports that Oracle rates neither flaw individually as exploitable without a login; the researcher’s claim concerns their combination.

The report lists affected Oracle Text releases as 19.3–19.32, 21.3–21.24 and 23.4.0–23.26.3, and affected Oracle REST Data Services releases as 24.2.0 through 26.2.1.

Updates and permissions to review

PostgreSQL administrators should update to the fixed release for their major version: 18.7, 17.12, 16.16, 15.20 or 14.25. PostgreSQL published those fixes on November 12.

Administrators of affected MySQL and Oracle deployments should apply the relevant patches. The Hacker News dates the MySQL fix to October 20. Oracle’s October patch advisory and November security patch advisory direct customers to product-specific patch availability and installation documents. Oracle recommends staying on actively supported versions and applying security patches without delay.

Database administrators should also review object-creation permissions and remove TEMPORARY privileges or broad temporary-table rights from accounts that do not need them, as the report recommends. This restriction does not address the Oracle chain, which reportedly starts without an account. ClickHouse operators should contact the vendor for a fix because no fixed version has been confirmed.

Read next

Sources

CVEs covered in this article

Back to home

Latest Cybersecurity News

All cybersecurity news →