AOMEI Backupper Update Hijack Exposes Windows Backdoor

·

Attackers Tampered with AOMEI Backupper's Update Servers for Minutes to Deliver a Backdoor

Attackers did not compromise AOMEI Backupper’s official update server for an extended period. Between May and July 2026, they changed files on the server for several minutes, causing Windows backup software that checked for updates during those windows to download and execute a malicious package. According to analysis by Google Threat Intelligence Group (GTIG), Mandiant, and ManageEngine’s Threat Analysis Lab, AOMEI fixed the server issue but did not guarantee that it would remove backdoors already installed on compromised systems. The incident showed how update-server tampering can create a brief intrusion window and exposed structural problems in a software supply chain where signed files alone cannot be trusted.

Turning a Legitimate Update Path into a Malicious Delivery Channel

AOMEI Backupper uses a background service to access AOMEI servers over HTTPS and check for updates and notification windows. This process is supported by two files, down_html.ini and Notifycfg.ini. down_html.ini specifies the update package to download and the checksum used for verification, while Notifycfg.ini determines what the notification service displays and executes.

The problem was that neither file was cryptographically verified. Although the checksum was intended to prove that the downloaded package was authentic, it was stored in the same unsigned files that attackers could modify. Consequently, the presence of a checksum did not guarantee that the update server and metadata could be trusted. Analysis found that no vulnerability in AOMEI Backupper or separate user manipulation was used. By changing the two text files on the server for several minutes, attackers turned the legitimate update process into an execution path for a malicious package. Programs that checked for updates during that period downloaded the package to their folders and then executed it according to the notification settings. Execution followed the permissions of the AOMEI process.

Three Attacks Detected Within Five to Ten Minutes

GTIG and Mandiant identified three periods during which the attacker tampered with the server. Files were modified for 10 minutes on May 28 from 05:40 to 05:50 UTC, for six minutes on July 16 from 08:33 to 08:39 UTC, and for five minutes on July 20 from 10:35 to 10:40 UTC. Only copies that checked for updates during the tampering windows were affected, while copies that checked immediately before or after those windows were confirmed to have received legitimate files. This behavior could enable attackers to evade time-based automated analysis and file scanning.

The attacks did not appear to target a particular country or organization. GTIG described the activity as opportunistic targeting in which systems that happened to check for updates at the relevant times became targets. The affected organizations identified through GTIG’s own telemetry spanned 11 countries in IT services, telecommunications, and higher education. The publicly named countries were Australia, Brazil, India, Pakistan, Portugal, Singapore, South Africa, Spain, Switzerland, Turkey, and the United States. The overall scope of the compromise has not been disclosed, but the actual number may have been higher than the confirmed figure. The report did not specify all affected versions of Backupper, while the version analyzed by ManageEngine’s Threat Analysis Lab was 8.3.0.

Multi-Stage Remote Control Using Radmin

On systems that received the malicious package, GTIG found that a downloader called CROSSROAD extracted files into a local Radmin folder. It then created a scheduled task that ran every three hours and deleted its own files from disk. The bundle included a legitimate, signed copy of Radmin, but it remained inactive on the analysis systems examined by ManageEngine. No listener, configuration, or new account was created, and actual remote control was handled by a separate implant.

When the scheduled task ran the legitimate Radmin program, it loaded an attacker-created DLL from the same location, and the DLL executed the SHINYROAD implant. SHINYROAD provided a remote shell and could receive and execute additional code on command. The communication destination was identified as a single hard-coded address with no backup address, and blocking that address could sever the attacker’s control communications. However, blocking the address does not automatically resolve the exposure of credentials through WebDAV.

CROSSROAD did not run identically on every system. It stopped execution when located in China, and ManageEngine also found that it stopped on systems with certain security agents installed. On some systems that passed these conditions, the open-source Adaptix C2 was additionally deployed on top of SHINYROAD and disguised as Nitro PDF software. AOMEI identified AnyViewer, a remote desktop tool that uses the same update flow, as a related target, but the publicly reported detections focused primarily on Backupper. GTIG tracks the attack group as UNC7017 and has not disclosed a country or known attacker affiliation.

A Path to Expose NetNTLMv2 Without Code Execution

The July 20 tampering also included a method that did not require execution of a separate program. The notification settings were changed to open an invisible one-pixel window that pointed to a WebDAV file on the attacker’s server. WebDAV allows remote files to be accessed as if they were on a local disk, and while Windows rendered the window, it could send the logged-in user’s NetNTLMv2 authentication response to the attacker’s address. This response is an authentication token that may be subjected to offline cracking or replay. Because there were few visible traces for users and no separate code execution occurred, information could already have been exposed even on systems where the backdoor was blocked.

Gaps in Update Integrity Verification

The central issue in this attack was not simply that the server was tampered with for a few minutes. Attackers could alter the update metadata, and clients followed the legitimate update process without cryptographically verifying that metadata. Organizations running AOMEI backup software should examine the times when updates were checked, the installation of Radmin, scheduled tasks running every three hours, SHINYROAD communications, WebDAV access, and unusual NetNTLMv2 traffic. AOMEI’s server-recovery report does not mean that the supply-side problem was fully resolved or that malicious persistence mechanisms remaining inside systems were removed. Unless the official update trust model adds signature verification and independent integrity checks, brief tampering with the update server can again turn a trusted software delivery path into a channel for remote control and credential theft.

Leave a Reply

Your email address will not be published. Required fields are marked *