Researchers at Kaspersky’s Global Research and Analysis Team (GReAT) have published new details on a sophisticated backdoor called GraphDoor, used to spy on government networks across Asia and Africa by hiding entirely inside traffic that looks like ordinary Microsoft and business-app activity. The findings, released November 17, 2026, form the second installment of Kaspersky’s ongoing “Operation ShadowCloud” research, and point — cautiously — toward possible Chinese-speaking threat actors behind the tooling.
Blending in With Business-as-Usual Traffic
What makes GraphDoor notable isn’t flashy exploitation — it’s the lengths the malware goes to in order to look boring. Once running on a victim machine, it opens a locally named pipe based on its own process ID and creates an innocuous-looking Intel_Log folder inside ProgramData, complete with a randomly generated subfolder. From there, it quietly gathers system fingerprint data: network adapter details, Windows version, public and private IP addresses, username, hostname, and a snapshot of running processes.
To make analysis harder, GraphDoor hashes the names of the Windows API functions it calls using the DJB2 algorithm, rather than referencing them directly, and uses a vectored exception handler — deliberately forcing a software error so that a small handler routine executes the real target function in its place. Collected data is encrypted with AES, Base64-encoded, and sent onward.
No Traditional Command Server Needed
Rather than reaching out to a conventional command-and-control domain, GraphDoor authenticates to login.microsoftonline.com as a malicious application — not a human user — using a client ID and secret embedded in the malware itself. That login hands it a legitimate access token for the Microsoft Graph API, which it then uses to upload stolen host data directly into a SharePoint document library, disguised as a file named main.<victim-code>.js.map — designed to look like a harmless JavaScript source map rather than stolen intelligence.
Each victim receives a random eight-character identifier, and Kaspersky observed the technique in use against SharePoint sites including buildersemp.sharepoint[.]com and cskautogo.sharepoint[.]com. From the same SharePoint lists, the malware can retrieve additional plug-ins on demand: a screenshot module, a process-listing module, and a keylogging module — though the keylogger is only fetched if a specific marker file already exists on the compromised machine, suggesting operators selectively escalate surveillance rather than deploying every capability by default.
This approach of tunneling command-and-control through trusted cloud services isn’t unique to GraphDoor. Kaspersky draws a direct comparison to Antino, a related Rust-based backdoor that performs a similar trick using Outlook and OneDrive instead of SharePoint. In both cases, network defenders monitoring for suspicious outbound traffic see only what appears to be routine calls to trusted Microsoft infrastructure.
A Modular Toolkit With Growing Capabilities
GraphDoor shares its command set with a sibling tool called GitDoor, while adding new features of its own — including a shell command that executes arbitrary text through cmd.exe, and an adsmload command that disguises payloads as fake MAC addresses encoded in Base64 before converting them back into executable bytes using a built-in Windows system call. Captured keystrokes and mouse activity are stored, XOR-encrypted with a random key, inside a file disguised within the Windows certificate cache directory.
Beyond spying tools, GraphDoor’s plug-in library also includes three separate network proxy modules that give operators a path deeper into a compromised network: a patched version of the open-source Erfrp reverse proxy that reads its configuration from memory instead of a file, a module built on the Ably real-time messaging platform’s official software development kit, and a SOCKS5 proxy that tunnels raw network traffic through Ably’s WebSocket-based messaging endpoint. Because command-and-control routing flows through these proxy layers or an Ably messaging channel rather than a hardcoded domain, defenders can’t simply extract a fixed indicator from a malware sample and block it outright.
Possible — Unconfirmed — Links to Known Espionage Groups
Kaspersky stopped short of formally attributing GraphDoor to a specific named group, but flagged circumstantial overlaps worth watching. Files tied to the related GitDoor tool appeared in the same folder as a malicious DLL linked by ESET to the China-linked group FamousSparrow, just a day apart. In a separate incident, GitDoor artifacts turned up alongside a loader for the PLEAD remote access trojan, historically tied to the group tracked as CopperTurtle, again within roughly 24 hours. Kaspersky cautioned that timing proximity alone doesn’t prove a direct operational link.
Separately, Sophos has previously observed the Ably messaging platform abused in an unrelated, China-linked intrusion, adding another data point to a pattern of East Asian threat actors experimenting with legitimate cloud messaging infrastructure for covert command-and-control.
Defensive Recommendations
Kaspersky’s report urges defenders to treat calls to Microsoft Graph, GitHub, and Ably as potentially suspicious rather than automatically trustworthy, specifically watching for unusual application sign-ins inside Microsoft Entra ID, unfamiliar SharePoint sites appearing in traffic logs, and document libraries containing oddly named files such as those ending in .js.map. Organizations in government and public-sector networks across Asia and Africa — the regions Kaspersky identified as current targets — should treat the appearance of unfamiliar remote-access tools, including legitimate ones like MeshCentral Agent, as worth investigating, and hunt specifically for sideloaded DLLs in folders such as PDF24 installation directories. Kaspersky’s published indicators of compromise, including file hashes, proxy server IP addresses, and the two Cobalt Strike-linked domains identified in the research, are available for defenders to add to existing blocklists and detection rules.
Leave a Reply
You must be logged in to post a comment.