NTLM Relay Mitigations Work Only When Every Authentication Path Uses Them
NTLM relay is an old attack with a durable operational lesson. An attacker who can intercept NTLM authentication can forward it to another service that accepts the same authentication without proving that the client intended to reach that service. The client's credentials may never be exposed, yet the attacker can act as the client.
Microsoft's Extended Protection for Authentication adds channel binding and service binding information to the authentication exchange. The server then checks that the authentication was intended for its service and, where TLS is used, for the same protection channel. The control is effective only when the client, server and protocol path all carry and validate the information.
Inventory the authentication paths
Start with services that accept Windows Integrated Authentication: IIS, Exchange, SQL Server, SMB, LDAP and custom applications using Negotiate. For each, record the protocol, TLS configuration, channel-binding setting, service-binding setting, signing requirement and the client population. A server that supports Extended Protection but is configured in a compatibility mode is not necessarily enforcing it.
LDAP deserves separate attention. LDAP signing and channel binding are different controls. Signing protects integrity, while channel binding ties authentication to the TLS channel. An LDAP client that connects without TLS cannot provide the channel binding material. Enabling the server setting without fixing clients can break authentication, so use audit mode and a staged rollout.
SMB signing should be required on sensitive servers and, where practical, on all SMB traffic. Disabling NTLM is stronger than protecting it, but it requires an inventory of legacy applications and devices. A policy that blocks NTLM outbound without an exception process may cause outages that lead operators to turn the control off. A policy that never blocks it leaves the relay path open.
Reduce the places an attacker can trigger authentication
Relay attacks often begin with a client being induced to authenticate to an attacker-controlled endpoint. WebDAV, printer discovery, shortcut files, email clients and browser authentication prompts can all create that trigger. Disable unnecessary services and restrict outbound SMB where the business does not require it. Block legacy protocols and require SMB signing for internal file shares.
Do not rely on a network location as proof of trust. A server in an internal VLAN can still be compromised. The authentication service should validate the target name and channel, not merely the source address. Certificate-based authentication and Kerberos should replace NTLM where the application supports them.
Verify the control in production
Test with a client that authenticates through a proxy or TLS terminator. Confirm that the server rejects a relayed authentication attempt and accepts the legitimate one. Check the event logs for the protection result. A successful login is not enough evidence if the server silently falls back to an unprotected mode.
After a certificate or load-balancer change, repeat the test. Channel binding depends on the TLS channel seen by the authentication endpoint. A reverse proxy that terminates TLS and forwards an unauthenticated connection can change what the server can verify. Document that path and keep a rollback procedure that does not disable the protection globally.
The goal is not to add one registry setting to every server. It is to make the entire authentication path prove its intended target, reduce the triggers that produce NTLM traffic and remove the protocol where modern alternatives are available.
Top comments (0)