This section covers some notes on the operational security of using Rubeus in an environment, with some technical examples comparing/contrasting some of its approaches to Mimikatz. The material here will be expanded in the future.
Any action you perform on a system is a detectable risk, especially when abusing functionality in "weird"/unintended ways. Rubeus (like any attacker toolset) can be detected in a number of methods, either from the host, network, or domain perspectives. I have a workmate who is fond of stating "everything is stealthy until someone is looking for it" - tools and techniques generally evade detection because either a) people are not sufficiently aware of the tool/technique and therefore not even looking, b) people can not collect and process the data needed at the appropriate scale, or c) the tool/technique blends with existing behavior to sufficiently sneak in with false positives in an environment. There is much more information on these steps and detection subversion in general in Matt Graeber and Lee Christensen’s Black Hat USA 2018 “Subverting Sysmon” talk and associated whitepaper.
From the host perspective, Rubeus can be caught during initial weaponization of the code itself, by an abnormal (non-lsass.exe) process issuing raw Kerberos port 88 traffic, through the use of sensitive APIs like LsaCallAuthenticationPackage(), or by abnormal tickets being present on the host (e.g. rc4_hmac use in tickets in a modern environment).
From a network or domain controller log perspective, since Rubeus implements many parts of the normal Kerberos protocol, the main detection method involves the use of rc4_hmac in Kerberos exchanges. Modern Windows domains (functional level 2008 and above) use AES encryption by default in normal Kerberos exchanges (with a few exceptions like inter-realm trust tickets). Using a rc4_hmac (NTLM) hash is used in a Kerberos exchange instead of a aes256_cts_hmac_sha1 (or aes128) key results in some signal that is detectable at the host level, network level (if Kerberos traffic is parsed), and domain controller event log level, sometimes known as "encryption downgrade".
One common way attack tools are detected is through the weaponization vector for the code. If Rubeus is run through PowerShell (this includes Empire) the standard PowerShell V5 protections all apply (deep script block logging, AMSI, etc.). If Rubeus is executed as a binary on disk, standard AV signature detection comes into play (part of why we do not release compiled versions of Rubeus, as brittle signatures are silly ; ). If Rubeus is used as a library then it's susceptible to whatever method the primary tool uses to get running. And if Rubeus is run through unmanaged assembly execution (like Cobalt Strike's
execute_assembly) cross-process code injection is performed and the CLR is loaded into a potentially non-.NET process, though this signal is present for the execution of any .NET code using this method.
Say we have elevated access on a machine and want to extract user credentials for reuse.
Mimikatz is the swiss army knife of credential extraction, with multiple options. The
sekurlsa::logonpasswordscommand will open up a read handle to LSASS, enumerate logon sessions present on the system, walk the default authentication packages for each logon session, and extract any reverseable password/credential material present. Sidenote: the
sekurlsa::ekeyscommand will enumerate ALL key types present for the Kerberos package.
Rubeus doesn't have any code to touch LSASS (and none is intended), so its functionality is limited to extracting Kerberos tickets through use of the LsaCallAuthenticationPackage() API. From a non-elevated standpoint, the session keys for TGTs are not returned (by default) so only service tickets extracted will be usable (the tgtdeleg command uses a Kekeo trick to get a usable TGT for the current user). If in a high-integrity context, a GetSystem equivalent utilizing token duplication is run to elevate to SYSTEM, and a fake logon application is registered with the LsaRegisterLogonProcess() API call. This allows for privileged enumeration and extraction of all tickets currently registered with LSA on the system, resulting in base64 encoded .kirbi's being output for later reuse.
Mimikatz can perform the same base64 .kirbi extraction with the following series of commands:
mimikatz # privilege::debug
mimikatz # token::elevate
mimikatz # standard::base64 /output:true
mimikatz # kerberos::list /export
Mimikatz can also carve tickets directly out of LSASS' memory with:
mimikatz # privilege::debug
mimikatz # standard::base64 /output:true
mimikatz # sekurlsa::tickets /export
As "everything is stealthy until someone is looking for it", it's arguable whether LSASS manipulation or ticket extraction via the LsaCallAuthenticationPackage() API call is more "stealthy". Due to Mimikatz' popularity, opening up a handle to LSASS and reading/writing its memory has become a big target for EDR detection and/or prevention. However, LsaCallAuthenticationPackage() is used by a fairly limited set of processes, and creating a fake logon application with LsaRegisterLogonProcess() is also fairly anomalous behavior. However full API level introspection and baselining appears to be a more difficult technical problem than LSASS protection.
Say we recover a user's rc4_hmac hash (NTLM) and want to reuse this credential to compromise an additional machine where the user account has privileged access.
Sidenote: pass-the-hash != over-pass-the-hash. The traditional pass-the-hash technique involves reusing a hash through the NTLMv1/NTLMv2 protocol, which doesn't touch Kerberos at all. The over-pass-the-hash approach was developed by Benjamin Delpy and Skip Duckwall (see their "Abusing Microsoft Kerberos - Sorry you guys don't get it" presentation for more information). This approach turns a hash/key (rc4_hmac, aes256_cts_hmac_sha1, etc.) for a domain-joined user into a fully-fledged ticket-granting-ticket (TGT).
sekurlsa::pthis used to over-pass-the-hash, Mimikatz first creates a new logon type 9 process with dummy credentials - this creates a new "sacrificial" logon session that doesn't interact with the current logon session. It then opens the LSASS process with the ability to write to process memory, and the supplied hash/key is then patched into the appropriate section for the associated logon session (in this case, the "sacrificial" logon session that was started). This causes the normal Kerberos authentication process to kick off as normal as if the user had normally logged on, turning the supplied hash into a fully-fledged TGT.
asktgtcommand is run (or Kekeo's equivalent), the raw Kerberos protocol is used to request a TGT, which is then applied to the current logon session if the
/pttflag is passed.
With the Mimikatz approach, administrative rights are needed as you are manipulating LSASS memory directly. As previously mentioned, Mimikatz' popularity has also led to this type of behavior (opening up a handle to LSASS and reading/writing its memory) being a big target for EDR detection and/or prevention. With the Rubeus/Kekeo approach, administrative rights are not needed as LSASS is not being touched. However, if the ticket is applied to the current logon session (with
/ptt), the TGT for the current logon session will be overwritten. This behavior can be avoided (with administrative access) by using the
/createnetonlycommand to create a sacrificial process/logon session, then using
/ptt /ticket:X /luid:0xa..with the newly created process LUID. If using Cobalt Strike, using the make_token command with dummy credentials and then kerberos_ticket_use with the ticket retrieved by Rubeus will let you apply the new TGT in a way that a) doesn't need administrative rights and b) doesn't stomp on the current logon session TGT.
It is our opinion that the LSASS manipulation approach is more likely (at the current moment) to be detected or mitigated due to the popularity of the technique. However the Rubeus approach does result in another piece of detectable behavior. Kerberos traffic to port 88 should normally only originate from lsass.exe - sending raw traffic of this type from an abnormal process could be detectable if the information can be gathered.
Sidenote: one way both approaches can potentially be caught is the previously mentioned "encryption downgrade" detection. To retrieve AES keys, use Mimikatz'
sekurlsa::ekeysmodule to return ALL Kerberos encryption keys (same with
lsadump::dcsync) which are better to use when trying to evade some detections.