ShadowCloud: which process uses your GitHub and SharePoint access?
Allowing a platform requires understanding the operation passing through it. ShadowCloud research helps teams review processes, identities and destinations with evidence they can correlate.
Ler em portuguêsBy Pentesty
The familiar platform participates in the chain
The November 17 SempreUpdate report follows Kaspersky GReAT's ShadowCloud research, published that day. Researchers identified victims mainly in government across Asia and Africa. They observed MeshAgent abuse, DLL sideloading and the GitDoor and GraphDoor backdoors. The entry point that allowed the agent to be installed remains unknown.
GitDoor uses GitHub repositories for commands and results. The second research installment describes GraphDoor authenticating an application and using Microsoft Graph and SharePoint; plug-ins also use Ably to forward traffic. These are observations from compromised systems. The publication establishes neither exploitation of the cloud platforms nor definitive attribution of the campaign.
Start with the process opening the connection
For a business investigation, propose a sequence starting at the workstation. Which process initiated communication? Where did its executable come from, which libraries did it load and what activity justified execution? Connect these elements with remote administration installations. Treat a familiar filename as one detail to check alongside its origin and behavior, with enough evidence to explain the finding.
Keep an inventory of remote agents, naming their owner, covered devices, authorized server and business purpose. Compare recent installations and configuration changes with maintenance records. When something differs, preserve evidence and establish scope before closing the alert. This approach supports investigating a legitimate tool's use without treating its mere presence as a diagnosis of infection.
Connect identity to the actual resource
For each relevant communication, try to identify the application, credential used and destination resource. In its service principal sign-in documentation, Microsoft describes authentication without a user and grouping by identity, status, IP and resource. Expand events needed for the timeline. Interpret an aggregated dashboard before using it as evidence of a particular session.
Document the site, repository or account actually accessed, as well as the provider's domain. Compare the destination with the workflow approved by its service owner and record what falls outside available visibility. The discussion of application identities and Microsoft 365 response helps organize containment and continuity decisions around the access actually involved.
State the scope of each log
GitHub describes its organization audit log as a record of activities affecting that organization. Do not assume your query explains operations in repositories controlled by third parties. Correlate available endpoint and network evidence, identifying the fields, times and destinations actually observed. Keep the distinction visible when presenting findings to the service owner.
Build a timeline with a source and time zone for each event. Distinguish authentication, authorization, reads and transfers that have actually been demonstrated. If inspection does not reveal content or a detailed destination, record that limitation. For a suspected credential, connect response to its corresponding identity, using the approach of verifying revocation at the provider.
Test containment alongside necessary work
In an authorized exercise, use a harmless file, a test identity and destinations controlled by the team. Agree beforehand which events should appear at the endpoint, identity layer and egress control. Compare collected evidence with that expectation. The aim is to assess a monitoring and containment decision without executing the implants described in the research or touching the reported infrastructure.
Also choose an approved workflow that must keep working, such as delivering an artifact to a corporate repository. Check whether the intervention stops the path defined as unauthorized while preserving the required task. Record the owner, observed result, gaps and review deadline. Pentesty recommends using this evidence together to accept the change and guide the environment's next assessment.
Frequently asked questions
Does a GitHub connection prove compromise?
It needs context: responsible process, identity, destination and operation. Compare those elements with approved use and record the evidence supporting investigation, without drawing a conclusion from the platform alone.
Do company logs cover every external destination?
Check each source’s scope before querying it. State which accounts and resources are visible to you and which operations require additional endpoint, network or resource-owner evidence.
Should every remote access tool be blocked?
Define which installations have an approved purpose and owner. Address discrepancies with evidence and a continuity plan, accounting for services that rely on remote administration to keep operating.
Is attribution necessary to contain a path?
A decision can start from demonstrated behavior and access in your environment. Preserve uncertainty about attribution and focus testing on permissions, destinations and results you can verify.
Sources and references
- SempreUpdate: GitDoor and GraphDoor news reference
- Kaspersky GReAT: ShadowCloud infection and GitDoor
- Kaspersky GReAT: GraphDoor, plug-ins and attribution limits
- Microsoft: service principal sign-in logs
- GitHub: organization audit log scope
Original Pentesty analysis, published November 17, 2026. The proposed scenarios are editorial guidance, with scope and validation to be defined for each environment.