Critical RCE Flaws Exposed in PostgreSQL, MySQL, Oracle & ClickHouse

Recent disclosure reveals dangerous remote code execution (RCE) vulnerabilities in major database systems—including PostgreSQL, MySQL, Oracle, and ClickHouse—that allow low-privilege or even unauthenticated attackers to run arbitrary code on database servers. These flaws exploit memory-corruption bugs embedded deep within common SQL engine code.

Security researcher Zheng Yu, working for Depthfirst, has published proofs-of-concept demonstrating full RCE in several top relational databases. Notably, some attacks require minimal permissions—just the ability to run SQL or create temporary tables—while Oracle’s findings include vulnerabilities exploitable without any login. None of these vulnerabilities have yet been observed in the wild.

Main Vulnerabilities and Scope

In PostgreSQL, the bug tracked as CVE-2026-116331 involves a scenario where a role with permissions to create database objects or use TEMPORARY privileges can exploit a type-confusion bug by constructing objects within its own temporary schema. This enables code execution with the operating-system user that runs the database server. The issue was patched in a security advisory on November 12. Affected versions include PostgreSQL releases prior to 18.7, 17.12, 16.16, 15.20, and 14.25.

MySQL and MySQL Cluster are vulnerable in versions 26.7.0 and 26.7.1 due to CVE-2026-101786. The flaw resides in the query optimizer, and although documentation suggests high privileges are required, proof-of-concept code showed that even accounts restricted to creating temporary tables—but without read access to other tables—can trigger the exploit. A patch was issued on October 20.

Oracle’s set of vulnerabilities includes CVE-2026-114819 in Oracle Text and CVE-2026-91609 covering the GraphQL interface in Oracle REST Data Services. The first was fixed in November; the second was addressed in October. Critically, the chain involving Oracle Text allows RCE without any login. Neither individual flaw was rated as exploitable without authentication according to Oracle’s advisories—but combined they enable this extreme exposure.

ClickHouse is also affected via an integer-overflow in the code that reads a binary row format. Attackers change a single setting—format_binary_max_string_size to zero—and can exploit the write bug to spawn a shell. This was demonstrated on version 26.7.5.10. No CVE has been assigned yet, and no fixed version is confirmed.

What System Operators Should Do

The most immediate step is installing vendor updates where patches are already available—for PostgreSQL, MySQL, and Oracle. For ClickHouse, users should contact the vendor to learn when a fix is expected. Beyond updates, it’s essential to audit what privileges database accounts truly need—not just what data they can read. TEMPORARY or broad rights for creating temporary tables are particularly risky and should be stripped from any role that doesn’t absolutely require them. Oracle’s vulnerable path remains open even without legitimate credentials.

These findings underscore a long-standing blind spot: database permissions models tend to focus on access to data tables, but often omit controls around the lower-level engine behavior that handles query parsing, planning, and execution in unsafe code layers.

While none of these vulnerabilities appear to be exploited yet, the potential for damage is high. If leveraged, they could allow attackers to take over database servers, access or modify sensitive data, and move laterally into wider infrastructure.

What this means: advocates for secure computing must re-examine how user roles are defined. Zero-or-least-privilege models need to account for what functions are exposed via SQL privileges—not just which tables are accessible. Database vendors should consider locking down temporary object creation privileges by default and hardening memory management in the underlying engine. Monitor for emerging patches and test upgrades carefully, especially in environments that host sensitive or regulated data. This discovery pushes the needle on what serious database security really demands.