Skip to content
Innovation & Growth

Attackers used GitHub and SharePoint as command channels, Kaspersky finds

Operation ShadowCloud hid its backdoor traffic in cloud services many firms allow. Victims were mainly government bodies.

W
WebPulse Newsroom
AI-assisted · 4 min read
Share on X LinkedIn
Attackers used GitHub and SharePoint as command channels, Kaspersky finds
In brief
  • Kaspersky says Operation ShadowCloud hides backdoor traffic in Microsoft SharePoint, GitHub and Ably. Victims were found in Asia and Africa, mainly in government.
  • The GitDoor backdoor takes commands from a public GitHub repository and reports results to a private one. Kaspersky says the traffic blends in with normal traffic and evades network-based detection.
  • Teams should ask which servers truly need GitHub access, and which remote-management tools run on their networks.

A trusted address is not a trusted conversation

Staff use GitHub and Microsoft SharePoint every day. Many teams allow them by default. Kaspersky's report describes a campaign that hides inside services like these.

Kaspersky says the traffic blends in with normal traffic. The implication, in WebPulse's analysis, is that a list of allowed domains will not catch it. Allowing a domain says who the traffic goes to. It does not say what the traffic is for.

What Kaspersky found

Kaspersky identified the campaign in early 2026 and named it Operation ShadowCloud. The attacker abuses MeshAgent, a legitimate remote-management tool, to drop two new backdoors. Kaspersky calls them GraphDoor and GitDoor, after the cloud platforms they use for control.

The traffic hides behind Microsoft SharePoint, GitHub and Ably, a messaging service. Kaspersky says this lets the implants blend in and evade network-based detection. Victims were in Asia and Africa, mostly in government.

Sophos touched on a fragment of the campaign in a recent brief. Kaspersky's report is part of a two-part blog. The text used here covers GitDoor in most detail, so this story does too.

How the entry works

In one case, Kaspersky found two copies of the MeshCentral Agent running inside a Taiwanese organisation. MeshCentral gives remote desktop, command-line and file access. The exact entry point is unknown. The evidence points to reuse of an earlier infection on the same network.

Once connected, the attackers dropped files that use side-loading. A genuine program loads a malicious file that has the name it expects. Kaspersky saw this with several legitimate programs, including an OpenJDK binary, Sandboxie's Start tool, a Python application and the Sysinternals Screen Magnifier tool. The program looks genuine, so the malicious code runs under its name.

The loader then reads an encrypted file called config.ini. It decrypts the contents and runs the backdoor in memory. Kaspersky says the backdoors also scramble their own logic and bury it in layers of wrapper code. That makes them harder for analysts to read.

A repository as a command desk

GitDoor's design is the part leaders should understand. Kaspersky says it uses two GitHub repositories. One is a public drop point for commands. The other is private, and it receives the results. Access to it needs a personal access token.

On first contact, the backdoor opens a new issue in the private repository. The issue number is then used in later requests for that machine. The backdoor then applies a "bug" label to the issue, or removes it if already present. Next it fetches the contents of the public repository. It repeats both steps until it receives an exit command. The source does not explain why the label is toggled.

Command files are named with the victim's unique code. The backdoor finds its own file, decodes it and runs the instruction. Kaspersky lists command types for keylogging, proxying traffic, file management and shellcode. It also says the backdoor downloads plug-ins for screen capture and process listing. Its first report goes to Ably as well as GitHub.

The source does not describe what defenders see on the network. One reasonable reading is that this looks like a developer's tool calling GitHub's interface, which would help the design hide.

Questions for your security team

The text available for this story does not say how many organisations were hit. It also does not say which of them used GitHub legitimately. Leaders can still test their own exposure.

First, ask which servers and workstations really need to reach GitHub, Ably or similar services. Second, ask for a list of remote-management tools such as MeshCentral, and who approved each one. Third, ask whether endpoint tools can flag a trusted program that loads a file from an unexpected folder.

Fourth, ask whether outbound traffic can be judged by behaviour. This is WebPulse's suggestion, not a Kaspersky finding: steady polling of one repository from a machine with no developers may be worth a look.

A tool should be trusted for what it does, not only for who hosts it. ShadowCloud shows why the difference matters.

Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: Kaspersky.