Operation ShadowCloud: New Backdoors Hide Inside GitHub and SharePoint Traffic
Kaspersky has detailed an active campaign, named Operation ShadowCloud, that targets government sectors in Asia and Africa. The company says it identified the activity in early 2026. The attackers abuse MeshAgent, a legitimate remote access tool, to drop two previously undocumented backdoors called GitDoor and GraphDoor.
What makes the campaign notable is how it hides. The backdoors send their command-and-control traffic through Microsoft SharePoint, GitHub and the Ably messaging service. Kaspersky's researchers say these cloud API calls are indistinguishable from legitimate enterprise usage, which lets the implants evade many network-based detections. The malware is also modular: it can download plugins on demand and run shellcode directly in memory, so new functions can be added without changing files on the infected machine.
GitDoor uses two GitHub repositories, one public for receiving commands and one private for uploading results. Their creation dates suggest preparation began at least by the end of 2024. GraphDoor authenticates to a malicious application via login.microsoftonline.com, then uses the Microsoft Graph API to stage data on a SharePoint drive. It creates a folder named "Intel_Log" in the Application Data folder, and loads plugins for screenshots, process listing and keylogging.
The starting point was a MeshCentral Agent on a Taiwanese organisation's network. How it was installed is unknown, though Kaspersky suspects the reuse of a prior infection. The actor is unidentified, but the activity overlaps with a China-nexus cluster Sophos tracks as Bronze Sandcastle, and shares tactics with FamousSparrow and CopperTurtle (BlackTech). Kaspersky says this does not prove a direct link.
Why it matters
Defences that rely on blocking or flagging suspicious destinations fail here, because the C2 traffic goes to GitHub, SharePoint and Microsoft login endpoints that most organisations allow. Security leaders must shift detection toward host behaviour and identity: unexpected remote management agents, unfamiliar app registrations, and unusual Graph API use. The likely reuse of an earlier infection also means a past compromise that was never fully cleaned up can be the entry point.
What to do
- Inventory and remove any MeshCentral or MeshAgent installs that were not deployed by your own IT team.
- Search endpoints for an "Intel_Log" folder inside the Application Data directory.
- Review Microsoft Entra app registrations and consent grants for unknown applications using Graph API access.
- Block or alert on outbound connections to the reported C2 address 27.102.66[.]162.