
Microsoft's change to stop Active Directory domain controllers from assuming RC4 is acceptable closes the cheapest route to Kerberoasting, but it does not close the route. What it leaves behind is narrower and easier to describe: accounts that someone explicitly told to keep RC4, service accounts whose passwords are weak, and domain controllers that are not yet enforcing. Those are the places a 2026 Active Directory assessment should spend its time, and the places a defender should be able to answer questions about.
TL;DR
- Per Microsoft, updates from January 13, 2026 added audit events, April 14, 2026 changed the default for accounts without an explicit setting to AES-SHA1 only, and July 2026 updates remove the rollback switch and programmatically enable enforcement.
- The vulnerability Microsoft tracks as CVE-2026-20833 is described as information disclosure: an attacker could obtain service tickets with weak or legacy encryption types such as RC4 and attack the service account password offline. Microsoft's page does not state a severity score.
- Enforcement removes a default, not an attack class. Any account that is explicitly configured to allow RC4 keeps the old exposure, and password strength still decides the outcome for AES-keyed accounts.
- Treat "no RC4 events in the log" as a starting point. Microsoft itself warns that missing audit events do not prove non-Windows devices will cope.
- A short Sigma-style rule for the leftover RC4 requests is included below. It was parsed and exercised on synthetic events, not on a live domain controller.
What Kerberoasting needs, and what RC4 contributed
Any authenticated domain user can ask a domain controller for a service ticket for any service principal name (SPN). The ticket is encrypted with a key derived from the service account's password. The MITRE ATT&CK entry for Kerberoasting (T1558.003) describes the consequence: if the ticket uses RC4 (encryption type 23), the hash can be brute-forced offline to recover the plaintext password, which may then support persistence, privilege escalation and lateral movement.
RC4 made this attractive for a historical reason. RFC 4757 documents the RC4-HMAC types Microsoft introduced with Windows 2000, and the RFC's own IESG note says it is Informational rather than a standard because of concerns about RC4 and MD4. The same document recommends against these types when aes256-cts-hmac-sha1-96 is available. For years, Active Directory kept using RC4 whenever an account did not declare what it supported.
How the defaults moved
Microsoft's detect and remediate RC4 usage in Kerberos page records the first step: updates on or after November 8, 2022 made AES-SHA1 the default for accounts where an encryption type was not explicitly set, while RC4 stayed the default for accounts that did not support AES-SHA1. It also says Windows Server 2025 domain controllers do not issue RC4 ticket-granting tickets.
The second step is in Microsoft's support article for KDC usage of RC4 and CVE-2026-20833. It lays out three phases:
- January 13, 2026, initial deployment. Updates provide audit events and introduce a temporary registry value,
RC4DefaultDisablementPhase. - April 14, 2026, enforcement with manual rollback. The Key Distribution Center (KDC) default for
DefaultDomainSupportedEncTypesbecomes AES-SHA1 only (0x18) for accounts with no explicitmsDS-SupportedEncryptionTypes. - July 2026, enforcement. Updates remove support for the rollback value and, in Microsoft's words, programmatically enable the enforcement phase.
One caution about that page: the metadata shows it was updated on September 17, 2026, yet the July phase is still written in the future tense. Do not infer from the text what your domain controllers run. Check patch levels. The Learn page above (last updated February 2026) likewise says only that Microsoft plans the change.
What is left for an attacker, and for a tester
Microsoft's guidance is explicit that administrators can still opt back in. If a service must keep accepting RC4 after April 2026, the support article recommends explicitly enabling RC4 in the msDS-SupportedEncryptionTypes bitmask for that service. That is a reasonable engineering decision for a legacy appliance, and it is also a list of accounts that retain the old exposure.
1. Accounts with an explicit RC4 setting
Enumerate service accounts that have an SPN and an explicit msDS-SupportedEncryptionTypes value that includes RC4. The Learn page notes that the attribute is stored in decimal and that processed values differ from stored ones: on Windows Server 2022 and earlier DES and RC4 are always shown, while from Windows Server 2025 only AES-SHA1 and stronger are. Read the stored attribute, not the effective one, or you will misreport.
2. Domain-wide overrides
DefaultDomainSupportedEncTypes is a KDC setting. Learn warns that changing it changes behaviour for every account without a value. A domain that "fixed" a broken application by widening this default has reintroduced the exposure for all accounts that have no explicit setting. Confirm the value on every domain controller, since one unpatched or overridden controller can answer requests the others refuse.
3. Passwords on AES-keyed accounts
Moving a ticket from RC4 to AES does not remove the offline step. MITRE lists AES as one mitigation and, separately, strong service-account passwords ("ideally 25+ characters") and Group Managed Service Accounts as others. Treating those as separate controls suggests that AES alone is not the whole answer. Our reading, which we have not measured here, is that AES raises the cost per guess for the attacker but that a short, human-chosen service password remains the weak point. Microsoft's group managed service account overview says the operating system manages the account password and the keys change periodically; it does not state a password length or rotation interval, so do not quote one.
The same page carries a compatibility warning that matters for migration: if a host is configured not to support RC4, authentication with an account that lacks AES always fails, and Microsoft advises configuring AES for managed service accounts.
4. Non-Windows services
The support article's note is blunt: the absence of audit events does not guarantee that non-Windows devices will accept Kerberos authentication after the April update, and it advises testing interoperability. Linux services and network appliances that authenticate with Kerberos are plausible first places to look; Microsoft's note does not name specific products. Each exception granted to keep one of them running should be recorded with an owner and an expiry.
Detection that survives the transition
MITRE's detection guidance for this technique points at service-ticket requests (Event ID 4769) using RC4 (0x17), unusual volumes of ticket requests from one account, and service accounts targeted outside their baseline. After enforcement, a genuine RC4 request should be rare, so each one is worth a look. Microsoft's Get-KerbEncryptionUsage.ps1 and List-AccountKeys.ps1 scripts, referenced from its Learn page and kept in the Kerberos-Crypto repository, help inventory what is in use; the repository notes its utilities are not explicit recommendations.
This Sigma-style rule flags RC4 service-ticket requests for non-machine, non-krbtgt services:
title: Kerberos service ticket issued with RC4 for a non-machine service
status: experimental
description: Flags Kerberos service ticket requests (4769) that used RC4-HMAC (0x17) for a service account, for review after RC4 default removal.
logsource:
product: windows
service: security
detection:
selection:
EventID: 4769
TicketEncryptionType: '0x17'
filter_machine:
ServiceName|endswith: '$'
filter_krbtgt:
ServiceName: krbtgt
condition: selection and not 1 of filter_*
falsepositives:
- Legacy or non-Windows services that still depend on RC4
level: medium
What was tested: the YAML was parsed with PyYAML in a sandbox, and the selection and filter logic was reproduced in a short Python script run over five synthetic events (one matching service-account request, one AES request, one machine account, one krbtgt, one non-4769 event). Only the first matched. The rule was not converted to a SIEM backend or run against real logs, so validate it in your environment before relying on it.
Where NTLM fits
Kerberos tightening is only half of the story, because a protocol downgrade is a way around either. Microsoft's roadmap for disabling NTLM by default (published January 29, 2026, with timelines marked as subject to change) describes enhanced NTLM auditing as available now, pre-release IAKerb and local KDC features in the second half of 2026 for Windows Server 2025 and Windows 11 24H2 and later, and network NTLM disabled by default in the next major Windows Server release. It states that NTLM stays in the operating system and can be re-enabled by policy. For testers, the practical point is the same as with RC4: the default changes, and the exceptions people configure to cope are what remains.
The PentStark view
This section is opinion. We think the most useful reframing for 2026 is that Kerberoasting has moved from a default-configuration finding to a configuration-debt finding. Before enforcement, a report could say "RC4 tickets are issued for service accounts" and be right for almost every domain. After enforcement, that sentence is only true where someone made an exception, so the finding becomes a list of named accounts, each with a reason and, ideally, an owner.
We would hedge in two places. First, we have not measured how much AES slows offline cracking in practice, so we would not promise a client that AES makes weak passwords safe. Second, we have read Microsoft's documents, not tested every domain controller build, so statements about what a particular environment enforces should come from checking that environment.
For programme owners, a sensible control to ask for is simple: a current inventory of SPN-bearing accounts, their encryption settings, their password age and type (managed or human-set), and the exceptions list. For assessors, ask for the same inventory, then compare it with what the directory actually says. That comparison tends to be more informative than the ticket requests themselves. It also connects to a theme from our recent posts on trusted publishing as an identity boundary and on agent and editor configuration as an execution surface: in each case a new default removes one weakness and leaves trust sitting in configuration that someone has to own.
What to do this quarter
- Confirm every domain controller has the 2026 updates and is in enforcement mode, rather than assuming it from the calendar.
- Export every SPN-bearing account with its stored
msDS-SupportedEncryptionTypesand review any that include RC4. - Check
DefaultDomainSupportedEncTypeson each controller for domain-wide overrides. - Move human-managed service passwords to Group Managed Service Accounts where the application supports it, and lengthen the rest.
- Track each non-Windows exception with an owner and an end date.
- Deploy a rule like the one above, tune it against your own logs, and alert on new RC4 requests.
References
- Detect and remediate RC4 usage in Kerberos (Microsoft Learn)
- How to manage Kerberos KDC usage of RC4 for service account ticket issuance changes related to CVE-2026-20833 (Microsoft Support)
- Advancing Windows security: Disabling NTLM by default (Windows IT Pro Blog)
- MITRE ATT&CK T1558.003, Steal or Forge Kerberos Tickets: Kerberoasting
- microsoft/Kerberos-Crypto repository
- RFC 4757, The RC4-HMAC Kerberos Encryption Types Used by Microsoft Windows
- RFC 8009, AES Encryption with HMAC-SHA2 for Kerberos 5
- Group managed service accounts overview (Microsoft Learn)
- PentStark: Trusted publishing moved the npm secret into an identity boundary
- PentStark: Your repo's AI-agent and editor config is now an execution surface
Last verified: 2026-11-06
