PostgreSQL patches 52 vulnerabilities including code-execution flaws in one batch
On November 12, 2026, PostgreSQL released 18.7, 17.12, 16.16, 15.20, 14.25, and 19 RC 2, a batch that closes 52 vulnerabilities — mostly type-confusion flaws leading to code execution — and marks the end of life of PostgreSQL 14. Apply the update, then reindex the affected indexes and take a fresh full backup before your next incremental backups.
November 12, 2026. The PostgreSQL Global Development Group ships an unusual maintenance batch: 18.7, 17.12, 16.16, 15.20, 14.25, and the second release candidate of PostgreSQL 19. 52 vulnerabilities and over 140 bugs are fixed in it. November 19, 2026. That is the planned general-availability date for PostgreSQL 19. Why it matters: this is the last release of PostgreSQL 14, now end-of-life, and a large share of the flaws are type confusions that execute arbitrary code from a mere low-privileged account.
52 CVEs, including a family of type confusions
This batch does not look like the usual quarterly fixes. The PostgreSQL Global Development Group normally publishes a handful of CVEs per cycle; here it closes 52 at once, accumulated over several months. The dominant pattern jumps out from the announcement: dozens of lines repeat the same wording — “type confusion executes arbitrary code” — almost always at CVSS 8.8.
The flagship is CVE-2026-6460: a type confusion after ALTER … SET DATA TYPE that executes arbitrary code. A crafted statement makes the engine interpret a value of one type as another, and the gap between the two in-memory representations turns into code execution. The flaw only requires a low-privileged account — PR:L in the CVSS metrics — making it a credible vector for anyone holding an application-level account, not just a superuser.
Others follow the same mechanism in widely used components: CVE-2026-18447 (output-parameter type confusion), CVE-2026-116304 (parallel RECORD use-after-free), CVE-2026-120139 (tuple-size integer wraparound). A run of use-after-free bugs hits the procedural languages: plperl (CVE-2026-116307, CVE-2026-116309), pltcl, and plpgsql — a reminder that functions written in interpreted languages widen the server’s attack surface.
Two families deserve extra attention because they touch backup and administration chains. CVE-2026-116318 enables SQL injection through the NOT NULL constraint name during pg_dump, and CVE-2026-116325 does the same through the index name in reindexdb. Most concerning for teams that restore archives received from third parties: CVE-2026-116314, code execution in pg_restore on a non-plain archive, which requires no prior authentication — the victim only has to launch the restore of a malicious file.
The end of life of PostgreSQL 14
This batch is also a lifecycle milestone: 14.25 is the last release of PostgreSQL 14, which now enters end-of-life. In practical terms, no further security or bug fixes will be published for that branch. Fleets still running PostgreSQL 14 — released in September 2021 — must plan a migration to a supported version, the project’s versioning policy guaranteeing five years of fixes per major branch.
This is a classic database trap: they run so well that nobody remembers to upgrade them. An end-of-life database concentrates risk: the next type-confusion CVE will simply not be fixed for PostgreSQL 14.
Post-update steps: reindex and rebase your backups
The update itself is cumulative and binary-only: no dump/reload or pg_upgrade, just stop the server and replace the binaries. But the batch demands post-update steps that many teams will miss, because they fall outside the usual “install and restart” reflex.
First, reindex. Several fixes concern index corruption: GIN indexes that could omit rows, pg_trgm GiST indexes that missed matches, and B-tree indexes on length-less bpchar columns. The instruction is explicit:
REINDEX INDEX your_index_name; -- suspected GIN or bpchar index
REINDEX TABLE your_table_name; -- affected table (also reindexes its TOAST) For pg_trgm, reindex the GiST indexes using the gist_trgm_ops operator class. The symptom to watch for is the missing chunk number error, a sign that a TOAST table was corrupted by an earlier REINDEX CONCURRENTLY.
Second, rebase incremental backups. Several fixes concern the generation of incremental backups, which could produce corrupt backups: the recommendation is to take a fresh full backup after updating and use it as the basis for subsequent incrementals.
Third, check NOT NULL NO INHERIT constraints on partitioned tables, now disallowed. The query provided in the announcement lists those to repair or drop:
SELECT conrelid::regclass, conname FROM pg_constraint
WHERE contype = 'n' AND connoinherit
AND conrelid IN (SELECT oid FROM pg_class WHERE relkind = 'p'); Verdict
If you run PostgreSQL in production, apply this batch in your next maintenance window: the density of type-confusion flaws exploitable by a low-privileged account (CVSS 8.8) makes waiting unjustifiable, and pg_restore adds a no-authentication vector to watch on any archive received from outside. The update does not end at restart: budget time to reindex the affected GIN, pg_trgm, TOAST, and bpchar indexes, and to take a fresh full backup if you rely on incremental backups. If you are still on PostgreSQL 14, treat this batch as the starting gun for your migration: the next CVE will no longer be fixed for you.