TL;DR
In early 2026, Kaspersky researchers uncovered an active espionage campaign named Operation ShadowCloud targeting government organizations in Asia and Africa. The attackers abuse the legitimate MeshCentral Agent and DLL sideloading to deploy two modular implants: GitDoor and GraphDoor. Furthermore, the Operation ShadowCloud GitDoor GraphDoor toolset hides its command-and-control traffic inside GitHub repositories, Microsoft SharePoint libraries, and Ably real-time messaging channels.
Delivery and Attribution
Investigators first identified two instances of the legitimate MeshCentral Agent running inside a Taiwanese organization, alongside additional government victims across Asia and Africa. Although the initial breach method remains unknown, evidence indicates the intruders reused a prior infection on the network to launch the remote administration agent. Once connected to its control server over port 443, the MeshCentral Agent dropped three files per instance: a trusted executable, a malicious loader DLL, and an encrypted payload file named config.ini.
To launch the loaders without triggering security alerts, the attackers relied on DLL sideloading. Specifically, they abused legitimate executables including OpenJDK (javaw.exe), Sandboxie Start (start.exe), a Python binary (pythonw.exe), and the Sysinternals Screen Magnifier utility (zoomit.exe), which they registered as a Windows service. These programs loaded malicious libraries named jli.dll, SbieDll.dll, python39.dll, WINMM.dll, and libcef.dll.
Confirmed Versus Suspected Attribution
Confirmed attribution to a single named threat group is not yet established, though external researchers at Sophos previously documented a small fragment of this activity. However, suspected attribution links Operation ShadowCloud to Chinese-speaking espionage groups, specifically FamousSparrow and CopperTurtle. On several victim hosts, those groups shared the exact same staging directories with GitDoor within a 24-hour window. For example, in March 2026, FamousSparrow deployed its implant in the %ProgramData%\excint\ folder just one day after GitDoor staged zoomit.exe in a subfolder at the same path. Similarly, in May 2026, CopperTurtle ran its PLEAD RAT loader from a PDF24 folder one day before GitDoor staged SbieDll.dll in that same directory. Furthermore, repository creation dates show the operators began preparing their GitHub accounts as early as September 2024.
Infection Chain
Kaspersky identified two distinct loader types that decrypt config.ini into memory. The Type 1 loader decodes the file using a custom Base64 alphabet, decrypts the first layer with a 16-byte RC4 key, and removes a second layer with a 16-byte XOR key. Next, to evade memory scanners, the loader converts the decrypted payload bytes into ASCII Ethernet MAC-address strings using RtlEthernetAddressToStringA and translates them back into raw binary bytes inside newly allocated memory via RtlEthernetStringToAddressA. A Type 1 loader variant uses FNV-1 API hashing and creates a local file-mapping object to load GraphDoor reflectively. Meanwhile, the Type 2 loader checks kernel libraries for a specific byte pattern, installs a vectored exception handler, zeroes the hardware breakpoint register CONTEXT.DR7, and applies a 16-byte XOR key to launch GitDoor.
Both backdoors use control flow flattening and nested indirect function wrappers to hinder static analysis. Internally, each implant splits its tasks across multiple C++ threads built around an ably client namespace.
Command-and-Control and Data-Exfiltration Behavior
In the Kaspersky Part 1 analysis of GitDoor, researchers detailed how the implant uses a hardcoded Personal Access Token to communicate with two separate GitHub repositories. It pulls encrypted commands and plug-ins from a public download repository and uploads stolen output to a private repository. To register a victim, GitDoor opens a new GitHub issue containing a session GUID and an eight-character victim code. It then encrypts host survey data with AES-256-ECB and XOR, saves the output as a fake JavaScript source map file named main..js.map, and uploads it to both the private GitHub repository (with the commit message “commit from INSOMNIA”) and an Ably messaging channel. Afterward, GitDoor repeatedly adds and removes a “bug” label on the GitHub issue while polling the command repository for task files.
GraphDoor and Modular COFF Plug-Ins
As detailed in the Kaspersky Part 2 report on GraphDoor, the second backdoor resolves Windows APIs using DJB2 hashes and triggers deliberate exceptions to run API calls inside a vectored exception handler. Instead of GitHub, GraphDoor authenticates to Microsoft Entra ID using a hardcoded client ID and secret to obtain an OAuth token for the Microsoft Graph API. It uploads its encrypted source map registration file to a compromised or actor-controlled SharePoint document library and queries SharePoint lists for victim-specific command items.
Both implants download and decrypt modular plug-ins and custom Common Object File Format (COFF) shellcodes directly in memory. Standard plug-ins include a screenshot tool (scr.ts), a process enumerator (pr.ts) that reports over a local named pipe, a file-relocation helper (moveDir.ts), and a keylogger (kl.ts) that activates when a log.lock file exists and writes XOR-encrypted keystrokes and mouse clicks to a certificate cache file. For network tunneling, the Operation ShadowCloud GitDoor GraphDoor framework reflectively loads Go binaries derived from the open-source Erfrp reverse proxy alongside two Ably proxy modules, one of which tunnels SOCKS5 traffic over Ably’s Pusher WebSocket protocol. Finally, operators can push custom COFF shellcodes to run WinPTY interactive shells, dump LSASS via a renamed Mimikatz build, steal browser cookies with CookieKatz, extract WhatsApp database keys, upload files to GoFile, or trigger PetitPotam NTLM relay attacks.
Defense and Detection Guidance
As Kaspersky researchers concluded, “By using cloud-based storage and messaging services, the actors blend in with normal traffic (as cloud API calls to Microsoft Graph, GitHub, and Ably are indistinguishable from legitimate enterprise usage, evading many network-based detections), and enable dynamic payload delivery, where on-demand plug-ins and shellcode are retrieved from cloud repositories, allowing for rapid updates and functional extensions without modifying host files.”
To defend against Operation ShadowCloud, security teams should treat any unapproved installation of remote administration utilities like MeshCentral Agent as a high-priority incident. Defenders should monitor endpoints for signed binaries like javaw.exe, start.exe, or zoomit.exe loading unsigned DLLs alongside config.ini files, and check for local named pipes or C:\ProgramData\Intel_Log\ directories. Finally, network administrators should inspect outbound traffic for workstation processes making repeated API calls to api.github.com, rest.ably.io, main.pusher.ably.net, or external SharePoint tenants, and block the known proxy server IP addresses and Cobalt Strike domains identified in the investigation.
Support Our Threat Intelligence
Find our threat intelligence and malware analysis helpful? Support our work today and unlock a 100% ad-free reading experience!