NTLM Relay Attacks Have Barely Changed Since 1993, and Windows Networks Still Can't Escape Them

NTLM relay is one of the oldest tricks in Windows network exploitation, and it is still working. The CVE-2026-124194 Outlook zero-day that CISA added to its Known Exploited Vulnerabilities catalog this month doesn't rely on a novel technique. It relies on the same structural flaw in NTLM authentication that researchers documented in the mid-1990s: a server can forward a client's authentication attempt to a second server and get authenticated as that client, because NTLM never binds a credential to a specific destination.
That gap has outlived three Windows server generations, two major Microsoft security pushes, and more patches than any single vulnerability class on the platform. Understanding why requires looking past the specific CVE and at the protocol underneath it.
The attack that refuses to die
NTLM relay works like this: an attacker coerces a victim machine into authenticating to an attacker-controlled listener, often through a crafted email, a malicious SMB share path, or — in the Outlook case — a calendar invite or embedded resource that triggers an automatic authentication handshake with no user click required. The attacker's listener captures the NTLM challenge-response exchange and immediately relays it to a real target server, such as a domain controller or an Exchange server, before the authentication session expires. The target server has no way to know the credential didn't originate from a direct connection, because NTLM was never designed to verify the channel it travels over.
Kerberos, the protocol Microsoft has pushed as the default since Windows 2000, solves this with mutual authentication and channel binding. NTLM predates that design and was kept around for backward compatibility with legacy applications, workgroup environments, and third-party software that never migrated off it. Twenty-six years later, that backward-compatibility decision is still the attack surface.
Why the 2026 Outlook bug is the 1997 bug with a new wrapper
What made CVE-2026-124194 notable wasn't a new relay technique — it was the delivery mechanism. Earlier NTLM-leaking bugs required a user to open a file or click a link. This one triggers during normal calendar sync processing, meaning Outlook clients leaked authentication material without any interaction. CISA's same-day KEV listing reflects how fast this kind of bug gets weaponized once proof-of-concept code circulates: coerced authentication plus relay to Active Directory Certificate Services (AD CS) or Exchange can hand an attacker a domain-admin-equivalent certificate within minutes of the first leaked hash.
The CVSS score of 6.5 looks almost reassuring next to that outcome, and that gap between the score and the real-world blast radius is exactly why NTLM relay chains keep getting underestimated inside patch-prioritization meetings. A “moderate” authentication-coercion bug that feeds into a privilege-escalation chain is not a moderate problem.
Microsoft shipped the fix years ago. Most networks haven't turned it on.
Microsoft has had tools to kill NTLM relay for years: SMB signing (mandatory since Windows Server 2022 defaults, optional before that), Extended Protection for Authentication (EPA) for IIS and AD CS endpoints, and LDAP channel binding for domain controllers. As of 2026, Windows also ships a tenant-wide "NTLM audit mode" that logs every NTLM authentication attempt without blocking it, specifically so administrators can see what would break before they disable NTLM outright.
Almost none of this ships enabled by default across an existing estate, because enabling it can break things that have quietly depended on NTLM for a decade: legacy line-of-business apps, print servers, older backup software, and third-party appliances that were never updated to support Kerberos or signed SMB. Flipping the switch without an audit period routinely breaks production systems, which is why most enterprise security teams treat "disable NTLM" as a multi-quarter project rather than a patch.
What actually closes the gap
The fix isn't a single patch — it's a sequencing problem. Security teams that have successfully retired NTLM exposure follow roughly the same order:
1. Turn on NTLM audit logging first, not last. Enable Group Policy's "Network Security: Restrict NTLM: Audit NTLM authentication in this domain" setting before touching anything else. This produces an inventory of every device and application still using NTLM, which is the information every later step depends on.
2. Enforce SMB signing and LDAP channel binding on domain controllers. These two settings close the most common relay-to-AD path and have shipped as secure-by-default in current Windows Server releases — but only for new installs, not upgraded ones.
3. Enable EPA on AD CS and Exchange endpoints specifically. These two services are the most common relay destinations because compromising either hands an attacker a path to impersonate any user or issue a certificate as one.
4. Use the audit data to retire legacy NTLM dependencies in order of exposure, not convenience. A single unpatched print server on an open VLAN is a higher-priority fix than a well-segmented legacy app with no internet-facing surface.
5. Only then move toward domain-wide NTLM denial. Skipping straight to this step is what causes the outages that make security teams gun-shy about the whole project in the first place.
The Outlook bug will get patched. The next NTLM-coercion vector will show up somewhere else — printer spoolers, file explorer previews, and calendar clients have all produced one in the last five years — because the underlying protocol gap hasn't moved. Networks that have already done the audit-and-harden sequence absorb the next one as a non-event. Networks that haven't will be reading about it in next year's KEV catalog too.