JupiteX Get the app
Science & Technology19 Nov 2026 · about 6 min

Researcher Demonstrates Database RCE Exploits Affecting PostgreSQL, MySQL and Oracle

The brief

Remote code execution means making a computer run code chosen by an attacker. Here, the target is the database server, so the result can go beyond reading or changing records: code runs on the machine hosting the database. The article says the PostgreSQL flaw can run code as the operating-system user that runs the database. Zheng Yu’s proof of concept for PostgreSQL began with a plain login role that could create database objects. It built objects in its own temporary schema without touching application data, then used a type-confusion bug to gain control of the server. The MySQL and ClickHouse research also reached code execution through database-engine bugs. Oracle’s route was different: a chain involving Oracle Text and Oracle REST Data Services could run commands through an exposed web endpoint before login. The article reports no real-world attacks, and says full attack chains were not independently reproduced. Operators should install fixes and review account capabilities.

01

What does remote code execution mean when the target is a database server?

Remote code execution means making a computer run code chosen by an attacker. Here, the target is the database server, so the result can go beyond reading or changing records: code runs on the machine hosting the database. The article says the PostgreSQL flaw can run code as the operating-system user that runs the database.

Zheng Yu’s proof of concept for PostgreSQL began with a plain login role that could create database objects. It built objects in its own temporary schema without touching application data, then used a type-confusion bug to gain control of the server. The MySQL and ClickHouse research also reached code execution through database-engine bugs.

Oracle’s route was different: a chain involving Oracle Text and Oracle REST Data Services could run commands through an exposed web endpoint before login. The article reports no real-world attacks, and says full attack chains were not independently reproduced. Operators should install fixes and review account capabilities.

02

How many database platforms were involved, and which disclosed flaws had vendor fixes?

The research covered four database platforms: PostgreSQL, MySQL, ClickHouse, and Oracle. Across them, the researcher described five flaws: one each in PostgreSQL, MySQL, and ClickHouse, plus two used together in the Oracle chain. The article says four flaws have vendor CVEs and fixes.

Those four are PostgreSQL CVE-2026-116331, MySQL CVE-2026-101786, Oracle Text CVE-2026-114819, and the Oracle REST Data Services GraphQL flaw CVE-2026-91609. The Oracle Text bug was patched in the November update; the GraphQL flaw was patched in October. PostgreSQL and MySQL also received fixes, though the article does not give their release dates.

ClickHouse is the exception: the researcher reported an integer-overflow write reached through binary-row input, but it has no CVE and no confirmed fixed version. The article advises operators to check with the vendor. None of the reported flaws has been seen in a real attack.

03

What could an attacker do with these flaws, and how was the Oracle attack path different?

The main consequence is that an attacker may turn ordinary database input into control of the server. In PostgreSQL, the article says code could run as the operating-system user running the database. The research also reports code execution against MySQL and ClickHouse, and command execution through the Oracle chain.

For PostgreSQL, a role able to create objects could build what it needed in a temporary schema, then exploit a type-confusion bug. The MySQL proof of concept used an account limited to creating temporary tables. For ClickHouse, the researcher set one format option to zero and reached an integer-overflow write while reading a binary row format.

Oracle differed because the reported chain needed no database account or login. It combined a bug in Oracle Text with a GraphQL flaw in Oracle REST Data Services, using an exposed web endpoint to reach engine code. The article reports no real attacks and says full chains were not independently reproduced. Oracle patched both flaws in separate updates.

04

Why can an account that cannot read application tables still reach vulnerable code in a database engine?

An account’s SQL permissions govern which tables and schemas it can access. But, as Zheng Yu argues, those permissions do not by themselves prevent the account from reaching the database engine’s C and C++ code. The engine runs code to read, plan, and execute queries, so even a restricted query can engage that machinery.

The article gives temporary-object privileges as a common route. PostgreSQL’s default TEMPORARY privilege, or permission to create temporary tables in MySQL and ClickHouse, lets an account create objects or send input that may trigger a memory bug. These actions do not require reading application tables. The PostgreSQL proof of concept built objects in its own temporary schema.

The distinction matters because reviewing only what accounts can read may miss what they can make the engine do. The research reports a PostgreSQL type-confusion bug, a MySQL optimizer flaw, and a ClickHouse integer-overflow write. Operators are advised to review capabilities as well as read access, remove unnecessary temporary privileges, and install vendor fixes.

05

How can permission to create temporary tables expose an account to bugs in the database engine?

Temporary tables are objects an account can create without needing access to existing application tables. The article says such rights can still give an account enough room to reach the database engine’s code for reading, planning, and executing queries. That code includes memory-unsafe C and C++ components, according to the researcher.

The examples show how the route differs by platform. PostgreSQL’s default TEMPORARY privilege let a plain login role build objects in its own temporary schema, then exploit a type-confusion bug. In MySQL, the proof of concept used an account that could create temporary tables but could not read existing ones. ClickHouse input could also reach a bug when a user set a format option to zero.

These are reported proofs of concept, not evidence of attacks in the wild. The article says operators should remove TEMPORARY privileges and broad temporary-table rights from accounts that do not need them. This reduces exposure to those routes, alongside installing available fixes, but does not address Oracle’s no-login chain.

06

What changes can operators make to reduce risk, and why would removing temporary-table privileges not block the Oracle path?

The article recommends looking beyond table-read permissions. Operators should review what accounts can do and remove PostgreSQL TEMPORARY privilege or broad temporary-table rights in MySQL and ClickHouse when accounts do not need them. Those permissions can give an account a way to send input that reaches vulnerable engine code.

Installing vendor updates is the other key step. The PostgreSQL, MySQL, and two Oracle flaws have vendor CVEs and fixes. ClickHouse has no CVE or confirmed fixed version in the report, so operators should check with its vendor. The article does not describe a fix date for PostgreSQL or MySQL.

Removing temporary-table rights cannot block the Oracle path because that chain needs no database account. It uses an exposed web endpoint, combining a bug in Oracle Text with a GraphQL flaw in Oracle REST Data Services to run commands before login. The article says Oracle patched the two flaws in its November and October updates. It reports no real attacks, but recommends addressing permissions and updates.

07

What is memory corruption in software, and how can it turn ordinary database input into code execution?

Memory corruption is a broad term for software errors that damage or misuse data stored in a program’s memory. The article does not give a general technical definition, but it describes the database engines’ C and C++ code as memory-unsafe. Such errors matter because the engine processes input while reading, planning, and executing ordinary SQL or other database formats.

The reported examples include a PostgreSQL type-confusion bug and a ClickHouse integer-overflow write while reading a binary row format. In MySQL, the researcher reached a query-optimizer flaw. These bugs let crafted objects or input reach vulnerable engine code, even when an account could not read application tables. The article does not detail the precise memory changes behind every exploit.

Zheng Yu reported proofs of concept that reached code execution on PostgreSQL, MySQL, and ClickHouse, plus an Oracle chain that ran commands before login. The article says full attack chains were not independently reproduced and none has been seen in a real attack. Vendor fixes cover four flaws; ClickHouse has no confirmed fixed version.

This brief was written by AI from the original reporting and checked by other models. Names, figures and quotes come from the source; read it for full context.

Read more in the JupiteX app

Pulse is free. New stories every 4 hours, each one broken into the questions that explain it.

Or read more news on the web