Security
Nov 17, 2026

PostgreSQL November 2026 CVEs: CVE-2026-6460 Leads 52 Fixes

A record 52-CVE security release, led by the CVE-2026-6460 type confusion RCE, is the last patch PostgreSQL 14 will receive. See what to patch this week.

Give me the TL;DR
PostgreSQL November 2026 CVEs: CVE-2026-6460 Leads 52 Fixes

On November 12, 2026, the PostgreSQL Global Development Group released PostgreSQL 18.7, 17.12, 16.16, 15.20, and 14.25, fixing 52 security vulnerabilities (CVEs). That is the most PostgreSQL has ever fixed in one release, nearly double August's record of 28. Thirty-eight are rated High and 32 score CVSS 8.8. The headline is CVE-2026-6460 (High, CVSS 3.1 8.8, AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H), a PostgreSQL type confusion (CWE-843) triggered by ALTER ... SET DATA TYPE that lets an authenticated role able to create objects run code as the operating system user that runs the database. It affects every version before 18.7, 17.12, 16.16, 15.20, and 14.25. Its root cause dates to 2004, and a researcher has already published crash and memory-leak steps and demonstrated full code execution on PostgreSQL 18.6.

Forty of the 52 affect PostgreSQL 14, and all 40 are fixed in 14.25. That is also the last release PostgreSQL 14 gets. The branch reached end of life the same day, and the 14.25 release notes say so directly: this is expected to be the last release in the 14.X series. The next scheduled PostgreSQL security release, on February 11, 2027, will fix CVEs that PostgreSQL 14 never receives.

Affected and unsupported? See Never-Ending Support (NES) for PostgreSQL.

TL;DR: what to patch this week

If you run PostgreSQL 15, 16, 17, or 18, update to the matching minor release (15.20, 16.16, 17.12, or 18.7). Rebuild or update every C-language extension at the same time, update client tools such as pg_dump and pg_restore on every machine that runs them, and work through the reindex steps in the release announcement.

If you run PostgreSQL 14, install 14.25 this week. It closes 40 of these 52. Then plan for February, because 14.25 is the last community fix you will get. In 2026 alone, PostgreSQL fixed 76 CVEs that affected version 14, and 46 of them scored 8.8. Next year's equivalents stay open.

If you run PostgreSQL 13 or older, none of these fixes were released for your branch. Neither were any of the other 44 PostgreSQL CVEs fixed earlier this year.

The November 2026 PostgreSQL CVEs at a glance

Read the privileges row carefully. Forty-seven of the 52 need an authenticated role, which sounds reassuring until you count the roles your applications, BI tools, migration jobs, extensions, and developers hold. Twenty-nine of the 32 scores of 8.8 need nothing more than a login that can create an object, and by default every login can create at least temporary objects.

CVE-2026-6460: the 22-year-old PostgreSQL type confusion

PostgreSQL lets you change the data type of a column or of a composite type's attribute. The problem is that values built under the old layout could survive that change: in prepared statements, in PL/pgSQL variables, in data dictionary entries, and, per the advisory, "many other parts of the system." Values were not self-identifying, so the server could later read old bytes as if they had the new type. ALTER ... SET DATA TYPE is the primary route, but not the only one; the advisory also lists ALTER TABLE ... ADD COLUMN; ROLLBACK; ALTER TABLE ... ADD COLUMN.

Researcher Mehmet Ince traces the primitive to May 5, 2004, when ALTER TABLE ... ALTER COLUMN ... TYPE was introduced. His code-execution technique depends on a later change, from May 14, 2015, that added expanded-object pointers. The CVE record credits 15 separate reporters, Ince among them. When that many people find the same bug independently, the bug class is not obscure anymore.

The fix is large. The 14.25 release notes link 28 commits for this one CVE, which add version tracking so the server can detect composite values that no longer match the type it is about to read them as, add locking around composite type changes, and tighten when ALTER COLUMN TYPE is allowed if other objects depend on the type.

The advisory calls the attacker an "object creator" and does not define the privileges that takes. Ince wrote that he could trigger the crash "as a lowest-privileged login role." Red Hat's assessment says an attacker needs a login that can create or modify objects. On PostgreSQL 14 specifically, every role can create objects in the public schema by default (PostgreSQL 15 removed that default), and in every version, every role can create temporary objects unless you revoke it.

Two operational notes the patch alone does not handle:

  • Rebuild your C extensions. Full protection requires rebuilding C-language extensions against the new minor release, and some extensions need source changes. The PostgreSQL RPM maintainers now require the November minor release for every compiled extension package. Extensions built against the new release that call DatumGetHeapTupleHeader() or heap_getattr() fail to load on older minor releases with an "undefined symbol" error.
  • Leaveposite_type_mismatch alone. This new server parameter can disable the new checks. The release notes say doing so carries security risks. If an extension breaks, fix or rebuild the extension.

Ince reported 21 cases in total, and six of this release's CVEs trace to his work: CVE-2026-6460, CVE-2026-116306, CVE-2026-120139, and CVE-2026-120148, all of which affect PostgreSQL 14, plus CVE-2026-116329 (18 only) and CVE-2026-120142 (17 and 18).

What else the November 2026 PostgreSQL release fixes

Fifty-two CVEs is a lot to reason about one at a time, so here they are in six groups. The full table, with scores and PostgreSQL 14 applicability for every CVE, is further down.

1. Memory corruption and type confusion in the PostgreSQL server (25 CVEs)

This is where most of the 8.8s live. Beyond the CVE-2026-6460 family, the type confusion bugs reach through SQL-language output parameters (CVE-2026-18447), min/max aggregate optimization (CVE-2026-116331), operator commutator and negator links (CVE-2026-116335), set-returning functions such as json_populate_recordset() (CVE-2026-120141), PL/pgSQL record fields (CVE-2026-120151), and REINDEX CONCURRENTLY with an index expression that drops the index (CVE-2026-116336).

Use-after-free bugs show up in parallel query coordination of anonymous RECORD types (CVE-2026-116304), portal setup (CVE-2026-116332), cursors over trigger transition tables (CVE-2026-120143), and domain checks on expanded records (CVE-2026-120150). Heap overflows hit text search: CVE-2026-116315 in tsvector_delete() and CVE-2026-116317 in ts_rank_cd(). An integer wraparound in tuple-size calculations (CVE-2026-120139) turns oversized tuple data into code execution. Five lower-scored bugs disclose server memory, including CVE-2026-116321 through an unterminated name value and CVE-2026-116308 through zero-length tokens in the dict_int module.

Eighteen of these 25 affect PostgreSQL 14. The other seven, including CVE-2026-116328, CVE-2026-116329, CVE-2026-120142, and CVE-2026-120144, affect only newer branches.

2. PL/Perl and PL/Tcl (6 CVEs, all affect PostgreSQL 14)

Five plperl bugs and one pltcl bug, every one an 8.8: use-after-free on reentrant prepared queries (CVE-2026-116307), reentrant return_next() calls (CVE-2026-116319), array and hash conversion (CVE-2026-116320), blessed-reference type confusion (CVE-2026-116309), tied-object exceptions that follow a stale sigjmp_buf (CVE-2026-116326), and reentrant result conversion in pltcl (CVE-2026-116323).

These are trusted languages, which means PostgreSQL grants USAGE on them to every role by default once installed. If plperl is installed and nobody uses it, every role on that database can still write a function in it.

3. PostgreSQL client tools: pg_restore, pg_dump, and friends (6 CVEs)

The server attacking the client is a recurring 2026 theme, and this batch continues it. A crafted non-plain-format archive can run code as the OS user running pg_restore (CVE-2026-116314, 8.8, affects 14). A compromised origin server can run code in the psql client that replays the restore.sql file from a tar-format pg_dump (CVE-2026-116330, 8.8, affects 14), which the advisory calls a gap in the fix for CVE-2025-8714. A compromised server can also feed pg_basebackup a backup manifest that it treats as a format string (CVE-2026-16240, affects 14).

The other three are newer-version only: SQL injection through pg_dump not-null constraint names (CVE-2026-116318, 18), through reindexdb index names (CVE-2026-116325, 17 and 18), and an untrusted search path in vacuumdb --missing-stats-only (CVE-2026-116313, 18), an invocation the pg_upgrade documentation recommends.

Client tool fixes only land where the client binaries get updated. That includes backup hosts, CI runners, and developer laptops, not just database servers.

4. PostgreSQL row security, views, and privilege checks (8 CVEs)

Individually lower-scored, collectively the bugs that undo an access model. INSERT ... ON CONFLICT DO UPDATE can read and modify every row in an auto-updatable view's base table regardless of the view's WHERE clause (CVE-2026-16237), and can read every row despite row security policies (CVE-2026-120154). DROP OWNED lets a role remove itself from restrictive row security policies on tables it does not own (CVE-2026-116311). REASSIGN OWNED skips the SET and CREATEDB checks that the equivalent ALTER ... OWNER commands enforce (CVE-2026-120146). A leaky function applied to an indirect expression can infer values a security_barrier view hides (CVE-2026-116327), and foreign keys and to_json() can invoke a cast function without EXECUTE privilege on it (CVE-2026-116303).

The highest-scored item in this group is CVE-2026-116316 (8.8): an object creator plants a domain, enum, or range type in pg_temp during a security-restricted operation, and a later query in the victim session that uses an unqualified type name runs the attacker's SQL as the victim. Seven of the eight affect PostgreSQL 14; the MERGE transition-table bug (CVE-2026-116312) does not, since 14 has no MERGE.

5. Extensions, foreign data wrappers, and logical replication (6 CVEs, all affect PostgreSQL 14)

Two bugs in extension scripts let a role with database-level CREATE run code as the bootstrap superuser: CVE-2026-6467 through schema renames and CREATE EXTENSION ... SCHEMA "$user", and CVE-2026-120153 through crafted schema or owner names, which the advisory compares to CVE-2023-39417. CVE-2026-120153 only matters if you install a non-bundled extension whose script uses the vulnerable substitution pattern. dblink has a use-after-free (CVE-2026-120147) and an sslkeylogfile connection parameter that appends to a file of the user's choosing (CVE-2026-14661; on PostgreSQL 14 this requires the server to be using libpq 18 or later).

A foreign server can overflow the heap by reporting more inserted rows than it was sent in a batch insert (CVE-2026-120149). And a compromised publisher can run code on a subscriber by naming a system catalog as a replication target (CVE-2026-120145, 8.8), though only superuser-owned subscriptions are vulnerable.

6. One pre-authentication bug, PostgreSQL 18 only (1 CVE)

CVE-2026-116322 is the only bug an unauthenticated client can trigger against the server. The client times a cancel request so that authentication_timeout fires while it holds a spinlock, forcing a server restart. It scores 5.3 because the impact is availability only, but CISA's assessment marks it automatable. PostgreSQL 14 through 17 are not affected.

Exploitation status of the November 2026 PostgreSQL CVEs

None of the 52 appear in the CISA Known Exploited Vulnerabilities catalog as of its November 13, 2026 update. CISA's SSVC assessment lists exploitation as "none" on every record it has scored so far, and the PostgreSQL project has not reported attacks. These CVEs were disclosed on November 12, 2026, so treat that as a snapshot.

CVE-2026-6460 is the exception worth watching. Ince's writeup publishes the SQL that crashes an unpatched server and a second step that leaks memory addresses, and includes a video of the full exploit running on PostgreSQL 18.6. It does not publish the SQL for the final command-execution step. He also ported the exploit to managed PostgreSQL platforms and privately shared reproducible proofs of concept with their security teams, most in August. A public crash and address leak, plus a demonstrated end-to-end chain, is a short distance from a public exploit.

And PostgreSQL bugs do get used in real intrusions. Rapid7 found that exploiting the BeyondTrust zero-day used in the December 2024 US Treasury breach required exploiting CVE-2025-1094, a PostgreSQL quoting flaw that had no CVE number at the time. We covered that history in our PostgreSQL 14 EOL post.

The bigger picture: 96 PostgreSQL CVEs in 2026

Here is the CVE count for each scheduled PostgreSQL security release over the last seven quarters, from the PostgreSQL security page:

All of 2025 produced 7 PostgreSQL CVEs. 2026 produced 96.

The reason shows up in the credits. Twenty-two of the 52 CVE records credit an AI security program or tool among the reporters: the Anthropic OSS program, Claude and Ada Logics, OpenAI Codex Security, and the OpenAI Security Research Team. Ince's writeup describes using an AI agent to learn PostgreSQL's type internals before he found CVE-2026-6460. PostgreSQL did not get worse at writing C. The cost of auditing it dropped, and more people and tools are doing it.

That cost does not go back up on February 11, 2027. It also does not check which branches are still supported.

PostgreSQL 13 is the preview. It reached end of life on November 13, 2025, at 13.23. In the year since, PostgreSQL has fixed 96 CVEs across four releases and shipped none of them for 13. The project does not assess unsupported branches, so no CVE record will tell you which of those 96 apply to 13. But 76 of them hit PostgreSQL 14, released one year after 13, and CVE-2026-6460's root cause predates both by more than 15 years. "Not assessed" means nobody checked. It does not mean clear.

That is PostgreSQL 14's position starting with the next release.

Which PostgreSQL versions are affected by the November 2026 CVEs?

PostgreSQL 15 reaches end of life on November 11, 2027, per the versioning policy. If you are planning a migration off 14 now, target 16, 17, or 18 so you are not repeating this exercise in a year.

PostgreSQL CVE mitigation guidance

The project's advisory and release notes offer no workaround for unpatched servers. What you can do is shrink how much of this attack surface an ordinary role can reach, which matters most once fixes stop arriving:

  • Remove default object creation. On PostgreSQL 14, run REVOKE CREATE ON SCHEMA public FROM PUBLIC; in each database. Most of the 8.8s in this batch need a role that can create objects, and Red Hat's CVE-2026-6460 assessment recommends the same revocation.
  • Review temporary object rights. REVOKE TEMPORARY ON DATABASE your_db FROM PUBLIC; blocks the pg_temp objects that the published CVE-2026-6460 steps and CVE-2026-116316 rely on. Test first: applications that use temp tables will need an explicit grant.
  • Drop languages and extensions you do not use. If nothing calls them, remove them with DROP EXTENSION. plperl, pltcl, and dblink together account for eight CVEs in this release alone.
  • Treat archives as code. Do not run pg_restore against archives from sources you do not control, and do not feed a tar-format dump's restore.sql to psql.
  • Avoid superuser-owned subscriptions to publishers you do not fully trust (CVE-2026-120145).

None of that addresses a memory corruption bug reachable by a role you have legitimately granted access. Only a patched binary does.

Related PostgreSQL CVEs

Several bugs in this release are direct relatives of earlier 2026 PostgreSQL CVEs that also affected PostgreSQL 14:

For the earlier releases, see our PostgreSQL 14 EOL post, which tables every PostgreSQL CVE fixed from February through August 2026. For support dates across every branch, see our PostgreSQL EOL dates timeline.

Every PostgreSQL CVE fixed on November 12, 2026

Scores are the PostgreSQL project's CVSS 3.1 base scores. NVD has not finished analyzing these CVEs as of publication.

Frequently asked questions

Is PostgreSQL 14 still supported?

No. PostgreSQL 14 reached end of life on November 12, 2026, and 14.25 is expected to be its last release. The PostgreSQL project says it will no longer receive security and bug fixes, so the February 11, 2027 release will not include a 14.x update.

Does CVE-2026-6460 affect PostgreSQL 14?

Yes. Every version before 18.7, 17.12, 16.16, 15.20, and 14.25 is affected. PostgreSQL 14.25 includes the fix, and you need to rebuild C-language extensions against it for full protection.

Is there a public exploit for CVE-2026-6460?

Partially. Mehmet Ince's writeup publishes the SQL that crashes an unpatched server and the SQL that leaks memory addresses, and shows a video of full code execution on PostgreSQL 18.6. It does not publish the final command-execution step. CVE-2026-6460 is not in the CISA KEV catalog.

Are managed PostgreSQL services affected?

Ince wrote that he had not seen an unaffected managed PostgreSQL vendor, including Postgres-compatible offerings, and privately shared reproducible proofs of concept with their security teams, most in August 2026. Check your provider's bulletin and reported minor version.

Do I need to reindex after updating?

Possibly. This release also fixes non-security bugs that could corrupt GIN indexes, pg_trgm GiST indexes, B-tree indexes on unbounded bpchar columns, and TOAST indexes rebuilt by REINDEX CONCURRENTLY. The release announcement lists the exact steps, along with a recommendation to take a new full backup if you use incremental backups.

Can I turn off the new composite type checks if an extension breaks?

The composite_type_mismatch parameter exists for that, but the PostgreSQL project says disabling the checks carries security risks. Rebuild the extension against the new minor release instead, and ask its maintainer if it needs source changes.

Is PostgreSQL 13 affected by these CVEs?

PostgreSQL does not assess unsupported branches, so no record will confirm it either way. PostgreSQL 13 received no fix for any of the 96 PostgreSQL CVEs disclosed in 2026, and CVE-2026-6460's root cause dates to 2004.

Taking action

PostgreSQL 14.25 closes 40 of these 52 CVEs. If you run PostgreSQL 14, install it this week, rebuild your extensions, and update your client tools. But understand what 14.25 is: the final community patch for a branch that collected 76 CVEs in 2026 alone, 46 of them at CVSS 8.8. The researchers and AI-assisted tooling that found them are not slowing down, and from February 11, 2027 onward, every fix they produce lands in 15 through 18 and stays open on 14.

If you can migrate to PostgreSQL 16, 17, or 18 before then, do it. If you cannot, NES for PostgreSQL is a drop-in replacement for community PostgreSQL 14 that keeps receiving security fixes after end of life, with FIPS-capable builds and no forced migration. Upgrade on your roadmap, not on a CVE's. Talk to us about coverage for your PostgreSQL estate.

Table of Contents
Author
Greg Allen
Chief Technology Officer
Open Source Insights Delivered Monthly