SilkParasite Cyber Espionage Operation Targets Central Asian Governments

A previously unreported cyber espionage operation, referred to as SilkParasite, has been detected targeting government bodies in Central Asia. This operation employs seven families of remote access tools (RATs), five of which have not been documented before: DriveSilkRAT, CookiETagRAT, NomadRAT, GoginRAT, and NodeEdgeRAT. Discovered in late 2025, SilkParasite is assessed to be associated with a China-nexus threat cluster with medium confidence.

Bitdefender Labs, which shared a technical report on the operation, noted that SilkParasite is intriguing due to the presence of AI-assisted development within its otherwise expert code. This differs from AI-generated malware, as SilkParasite's tools exhibit characteristics typical of professional espionage developed by human operators, with AI likely used to enhance efficiency.

The report highlights a phishing lure that appears to be AI-generated, marking a rare instance of sloppiness that may have been intentionally designed to mislead attribution efforts. SilkParasite is the third significant threat actor to target Central Asia in recent years, following UAC-0063 and FamousSparrow. A notable connection to China is established through the use of a backdoor named BLOODALCHEMY, which is an updated version of Deed RAT, itself a successor to ShadowPad. Both ShadowPad and PlugX are commonly utilized by Chinese hacking groups.

BLOODALCHEMY was first documented by Elastic Security Labs in October 2023 and was observed in attacks against government organizations in Southern and Southeast Asia. This C-based backdoor is part of a broader toolset launched via a DLL loader sideloaded using a legitimate binary. It supports basic commands for gathering host information, overwriting malware binaries, and uninstalling itself.

Another indicator of the operation's Chinese connection is the updated version of SpiceRAT, which can download and execute binaries and commands. This tool is attributed to another Chinese-speaking threat actor known as SneakyChef. Attack chains typically begin with password-protected RAR archives containing malicious Microsoft Office documents, delivered through spear-phishing emails. The password is provided in the email body, and opening the document triggers a macro that initiates a DLL sideloading sequence to deploy the first-stage payload.

Bitdefender noted that the phishing lures were tailored to specific regions, with recovered documents designed to appear relevant to government entities in Uzbekistan, Turkmenistan, Kyrgyzstan, Tajikistan, and Kazakhstan, some impersonating specific ministries. Additionally, a document found on a public malware-sharing platform was addressed to a Georgian government entity.

Significantly, the macro checks for the presence of Kaspersky antivirus software before execution, indicating attempts to evade detection given the software's prevalence in the region. The tools used in the attack feature a plugin-oriented architecture, allowing operators to expand capabilities while minimizing detection footprints. This modular system enables threat actors to upgrade components without replacing the underlying framework.

The seven malware families span four programming languages: .NET, C++, Go, and JavaScript, with DLL sideloading as the primary delivery method. This approach involves using a legitimately signed program to run malicious code placed under a name that the executable expects. Bitdefender reported observing approximately 65 instances of DriveSilkRAT, primarily in the Asia region.

Indicators of AI assistance are evident in GoginRAT, which includes Go test functions and a hard-coded AES key, and NodeEdgeRAT, which has a configuration field for an encryption key labeled "change_this_key." Both NomadRAT and GoginRAT share a similar architecture, suggesting a high-level design implemented in two languages, potentially scaled using AI-assisted workflows.

The most consistent detection method across the campaign is DLL sideloading, with reliable signals arising from the pairing of a legitimate application loading a library placed beside it while running from an unusual location. More broadly, low-footprint plugin-based implants operating through legitimate cloud services are poorly served by volume-based detection, necessitating behavioral baselines to flag unusual relationships between processes and network services rather than relying on signatures for individual artifacts.