Kerberos RC4 in 2026: Are Your Service Accounts Ready?

Hiya folks,

You probably know the sort of account I’m talking about. Something like svc-reports, created years ago so a reporting server could talk to SQL. The password was set once, stored somewhere (hopefully), and nobody has touched it since. The person who created it left three reorganisations ago. It just works.

Or it did. Microsoft changed how domain controllers handle older Kerberos encryption in 2026. If your DCs are up to date, that forgotten service account might already be relying on an exception, or an application might be struggling to authenticate.

Here’s what changed, how to spot a problem and what to do without breaking the thing that has been quietly running for years.

What changed, and why should we care?

A quick Kerberos refresher: when an application connects to something like SQL Server, it can ask a domain controller for a service ticket. Think of it as a pass that lets the application prove it can connect to that service. Part of the ticket is protected using a key belonging to the service’s account.

The problem is RC4, an older encryption option. An attacker who can request a service ticket may be able to take it away and try to crack the service account password offline. That’s called Kerberoasting. AES is the stronger alternative. It makes that attack harder, although a weak password is still a weak password.

Microsoft rolled out changes for CVE-2026-20833 in stages:

  • January 2026: updated domain controllers began logging warnings about potential RC4 problems.
  • April 2026: where no explicit encryption setting overrides it, the domain controller began assuming AES rather than RC4 for service-ticket requests.
  • July 2026: Microsoft removed the temporary rollback setting from updated domain controllers.

One important distinction: RC4 hasn’t been removed altogether. Explicit settings on a service account, or a domain controller’s configured default, can still permit it. So an application working today doesn’t necessarily mean it’s using AES.

Which service accounts might be affected?

There are a few ways older service accounts and applications can run into trouble:

  • An account without AES keys. Very old service-account passwords are worth checking. Changing the password can generate the missing keys, but password age alone doesn’t prove they’re absent.
  • An older application or keytab. Some clients can only use RC4. A Linux or Java service might have a keytab (a file of Kerberos keys) containing only RC4 keys. The DC may issue an AES ticket successfully, but the application can’t use it.
  • An explicit RC4 exception. Someone may have allowed RC4 on a particular account, or changed the domain controller’s default to keep a legacy integration alive. The latter affects every account relying on that default.

That last one is worth looking for even when nothing appears broken.

How can we check?

You need the ActiveDirectory PowerShell module for the account query, and permission to read domain controller event logs for the log checks. The examples below are read-only, haven’t been lab-tested for this post, and should be run in an appropriate admin environment. No passwords or domain settings are changed.

1. Find user-based service accounts

Get-ADUser -LDAPFilter '(servicePrincipalName=*)' `
  -Properties 'msDS-SupportedEncryptionTypes',PasswordLastSet |
  Select-Object SamAccountName,PasswordLastSet,
    @{Name='EncTypes';Expression={$_.'msDS-SupportedEncryptionTypes'}}

A service principal name (SPN) identifies a service in Kerberos. This query finds user accounts with SPNs in the domain you’re connected to, not every service identity. Computer accounts, gMSAs, other domains and accounts without SPNs are outside its scope.

Look for old passwords and accounts with explicit encryption settings. A blank EncTypes means the account relies on a default; it doesn’t mean RC4 is being used, or that AES keys are present.

2. Check what the domain controllers are saying

Run this on each relevant domain controller:

Get-WinEvent -FilterHashtable @{LogName='System'; Id=201..209} -MaxEvents 300 |
  Select-Object TimeCreated,Id,ProviderName,Message

Look for events from Kdcsvc, particularly 205, which warns about an insecure explicit domain-controller default. Other events in that range can point to missing keys or clients that only offer older encryption. Microsoft’s KB explains each event, so you don’t have to memorise the numbers.

No events? Good to know, but not proof everything works. Logs may have rolled over, and an application with an old keytab can still fail after a ticket is issued.

3. Check actual ticket usage

For a closer look, Microsoft provides two scripts in its Kerberos-Crypto repository. Follow the Microsoft Learn instructions for running them with appropriate permissions on a DC:

.\List-AccountKeys.ps1
.\Get-KerbEncryptionUsage.ps1 -Encryption RC4 -EncryptionUsage Ticket

The first reports available keys for accounts seen in the logs. The second looks for RC4-encrypted tickets, rather than merely RC4 session keys. By default, the scripts examine the local DC’s available Security events from the previous 30 days. Ensure Kerberos auditing is enabled and check other DCs if needed.

Found something? Don’t just reset the password

Find the owner first. What uses the account? Who can test the application? If nobody knows, that’s already a useful finding.

If AES keys are missing, plan a password reset with the application owner. Anything holding the old password or a keytab may need updating, otherwise your security fix becomes tomorrow morning’s outage. For an RC4-only client, check whether the application supports AES or needs an update. Regenerate keytabs when required.

A group managed service account (gMSA) may be a better long-term fit if the application supports it. AD manages its password rotation. I wrote about setting one up for an AD connector a while back.

If you genuinely need RC4 for a while, treat it as a documented, time-limited exception for the affected account. Don’t re-enable it as the domain-wide default just to make the noise stop. An exception with no end date is just the old default in a fake moustache.

I touched on service accounts in Non-Human Identity 101. They were non-human identities long before that term got fashionable. This is a good reminder that an identity needs an owner and a lifecycle, even when it’s been quietly working since 2011.

Found an old appliance or keytab that refused to cooperate? I’d be interested to hear what turned up.

Catch ya in the next one.

References

Leave a comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.