AhsayCBS vulnerabilities exploited against backup infrastructure
AhsayCBS vulnerabilities were exploited to bypass authentication, run commands and install malware on systems used to hold backups, according to cybersecurity researchers at Huntress. The activity began in early October 2026 and used CVE-2026-105133 and CVE-2026-105134, two flaws that Ahsay Backup patched in mid-October.
The vulnerabilities are assessed to have been used as zero-days, meaning attackers exploited them before fixes were available. Huntress first saw a rapid attack pattern involving web shells, which are small server-side programs that let an intruder issue commands, and XMRig cryptocurrency miners. The later intrusion was more extensive. The operator placed shells in several locations, ran two miners, prepared tunneling tools and installed a custom implant on servers responsible for backups.
The attacker left five types of Jakarta Server Pages web shells. JSP, formerly called JavaServer Pages, is a technology that allows a web server to run Java-based page code. One shell, named nsmind.jsp, provided a signed interface for command execution and file transfers through parameters called exec, cd, upload and download. AhsayCBS.jsp acted as a loader, hiding encrypted payloads in PNG image pixel data and running them from memory.
Two other shells used a hard-coded key before accepting commands. One ran commands through cmd.exe, the Windows command interpreter. The other could also write Base64-decoded data to any location reachable by the service account, allowing the attacker to deploy additional tools. The fifth family used names such as status_.jsp or st_.jsp. These shells could run commands, recursively delete paths supplied as input and remove themselves with cmd, del and rm parameters.
Huntress said AhsayCBS.jsp was disguised as an internal image preview service, including a comment block describing a watermark endpoint. The shell relied only on JSP features available inside the platform and fully qualified Java class names, so it could compile on any servlet container and on Java Development Kit 6 or later. A servlet container is software that runs Java web applications, while the Java Development Kit is the software package used to build and run Java programs.
The intrusion deployed one XMRig miner while making it appear to be Microsoft Edge. A second miner, named kvm-agent.exe, used a different cryptocurrency wallet and mining pool. It was configured to restart every minute through a SYSTEM scheduled task, giving it a way to return after termination.
The most distinctive payload was a Python backdoor called main.py. A backdoor is hidden software that gives an intruder continued access. The program was designed to run only on the computer where it had been staged and to avoid sandboxes or analysts' machines. It calculated the SHA-256 cryptographic hash of the host's MachineGuid, a Windows identifier for that computer, then searched the implant's directory for a marker file with the same name. If the expected condition was met, it loaded an embedded x64 DLL directly into memory and contacted the command-and-control server creepexplosiondarknight[.]com. A command-and-control server is an attacker-controlled system that sends instructions to compromised devices.
The attackers also configured Microsoft Defender exclusions, which prevent selected files or folders from being scanned, and supplied GOST and Chisel tunneling tools. Tunneling tools create a connection that can carry traffic through another route. Legitimate remote-access software, including AnyDesk, Atera and Splashtop, was also delivered to help maintain access.
Evidence suggested that backup servers were the deliberate target. On a related Veeam server in the same environment, Huntress observed a command consistent with NetExec's Veeam credential-theft module. A PowerShell script, apparently launched remotely through Windows Management Instrumentation, queried the Veeam backup relational database used by many web applications for stored credentials and decrypted them with DPAPI. Those credentials may provide access to hypervisors, repositories and domain accounts. Huntress could not determine how many credentials were retrieved, but said the AhsayCBS access combined with the Veeam query indicated a specific interest in backup infrastructure.
Ahsay Backup later released version 10.3.6.1 with additional security hardening updates and urged partners to apply the latest fixes. The company said cybercriminals are increasingly using artificial intelligence powered techniques to accelerate and improve vulnerability exploitation, describing that as an industry-wide challenge rather than an issue unique to Ahsay.
In a status update, Ahsay said partners began reporting unauthorized access to AhsayCBS server environments on October 9, 2026. The company released hotfixes for versions 9 and 10 on October 10 and 12. Customers can look for suspicious folders created at the root of a hard drive, including ahsay_ts, ahsay_t8, ahsay_t9, ahsay_t20, ahsay_t21, srv, user and opt. Other warning signs include multiple randomly named JSP files in the %CBS%\\webapps folder, folders made from lowercase letters and numbers such as xtj7iu, kje74x and 103y9r, unauthorized RPS receivers, and AhsayCBS accounts with Admin privileges.
If malicious activity is found, Ahsay advises reinstalling Windows or Linux on the backup server, installing the latest AhsayCBS version and restoring the server configuration from Auto Save. The selected configuration version should predate the compromise, ideally by at least one day before the timestamp of any suspicious folder or file created during the unauthorized access. For teams reviewing their wider exposure, AEU-I describes its service as security-first IT, infrastructure and consulting.