Home / Databases / SQL Server

DatabasesRunbookIntermediate

SQL Server October 2026 security updates (CVE-2026-96956 and 21 more) - which build to install, known issues to read first, and a safe rollout

Microsoft's 13 October 2026 updates for SQL Server 2017-2025 fix an elevation of privilege flaw (CVE-2026-96956) among 22 CVEs in the SQL Server 2025 update. Two known issues carried over from September affect MSDASQL linked servers and sys.dm_exec_requests during recovery. Check them, pick the right package, and patch availability groups in the right order.

Deepak Bhardwaj3 min read

Applies toSQL Server 2017, 2019, 2022 and 2025 on Windows and Linux; GDR and CU servicing branches; Always On availability groups and failover cluster instances

What's in this round

Released 13 October 2026 for SQL Server 2017, 2019, 2022 and 2025. For SQL Server 2025 the package is KB5126967 (CU9 + GDR), build 17.0.5015.4, which fixes 22 CVEs:

Type Count Examples
Remote code execution 9 CVE-2026-94512, CVE-2026-83945, CVE-2026-70189
Elevation of privilege 7 CVE-2026-96956, CVE-2026-88088, CVE-2026-85919
Information disclosure 4 CVE-2026-88086, CVE-2026-88081
Denial of service 2 CVE-2026-94520, CVE-2026-67626

For other versions, find the October 2026 KB for your branch on the SQL Server update centre page. Each KB lists the exact build.

Known issues to read before installing

The SQL Server 2025 KB lists two known issues (the same ones appeared in the September 2026 updates):

  1. MSDASQL linked server queries fail with error 7416 when the linked server uses a provider string (@provstr). See MSDASQL linked server error 7416.
  2. Access violation when querying sys.dm_exec_requests while a database is recovering, which creates a dump file. See sys.dm_exec_requests access violation.

If you use MSDASQL linked servers (ODBC-based links to other databases) or run monitoring that polls sys.dm_exec_requests (most monitoring tools do), test those specifically before production.

1. Pick the right package

SQL Server servicing has two branches:

Your server Install
On the GDR branch (security fixes only, no CUs installed) The GDR package for your version
On a CU (any cumulative update installed) The CU + GDR package (for 2025: KB5126967)

Check where you are:

SELECT SERVERPROPERTY('ProductVersion')     AS build,
       SERVERPROPERTY('ProductUpdateLevel') AS cu_level,
       SERVERPROPERTY('ProductUpdateReference') AS kb;

Once you've installed a CU, you can't go back to the GDR branch without uninstalling.

2. Pre-checks

  • Full backups (and log backups) completed and restorable.
  • Enough free disk on the system drive for the installer.
  • For AGs: all replicas healthy and synchronised; note the current primary.
  • Agent jobs that run during the window paused or moved.
  • Test server patched first with your monitoring tool connected (it exercises the DMV issue).

3. Patch an availability group

  1. Patch secondary replicas first, one at a time. Set them to manual failover if your AG uses automatic failover with only two synchronous replicas.
  2. Confirm each secondary is back and synchronised:
SELECT ar.replica_server_name, drs.database_id, drs.synchronization_state_desc, drs.synchronization_health_desc
FROM sys.dm_hadr_database_replica_states drs
JOIN sys.availability_replicas ar ON ar.replica_id = drs.replica_id;
  1. Fail over to a patched synchronous secondary.
  2. Patch the old primary.
  3. Optionally fail back; restore automatic failover settings.

Avoid querying sys.dm_exec_requests for databases in recovery during the failover window; that's when the known access violation can fire.

4. After installing

SELECT @@VERSION;
SELECT name, state_desc FROM sys.databases WHERE state_desc <> 'ONLINE';
  • Check the SQL Server error log and the \Log\ folder for new dump files.
  • Run a test query through each linked server.
  • Confirm Agent jobs ran on schedule after the window.

Checklist

  • Correct package for GDR or CU branch identified
  • Known issues reviewed against MSDASQL and DMV use
  • Patched on a test server with monitoring connected
  • AG secondaries first, then failover, then old primary
  • Build, database state, dump files and linked servers checked afterwards

Sources