Description

At the core of an advanced malvertising attack scheme known as FakeAgent, SectopRAT was utilized by the cybercriminals in exploiting Anthropic’s Claude platform for the distribution of a covert remote access trojan that was able to steal sensitive information from the target system. Victims searching Bing for the Claude desktop app were directed through sponsored results to a public Claude Artifact, which ultimately redirected them to a fake download site serving a trojanized ClaudeDesktop.exe. Although Anthropic removed the malicious artifact on July 22 after Huntress reported it, the installer had already accumulated nearly 7,100 views, contributing to the compromise of at least 29 organizations. Rather than using Anthropic’s legitimate installer, the attackers repurposed JetBrains jcef_helper.exe to enable DLL sideloading and executed additional malware through a signed IBM SPSS binary. The campaign also employed EtherHiding, embedding command and control (C2) configuration within blockchain transactions, allowing operators to update infrastructure without modifying malware binaries. To evade detection, the malware performed extensive anti-virtualization checks, inspected GPU hardware characteristics, and even relied on DirectX compute shaders to execute AES 256 CTR decryption on the GPU instead of conventional CPU based cryptographic APIs, significantly complicating malware analysis. Huntress analysts, assisted by Claude Opus, extracted the DirectX shader, reconstructed the custom encryption routine, and ultimately recovered the hidden payload after correcting a single transcription error in the encryption key. Analysis revealed the malware belonged to the SectopRAT (ArechClient2) family, a .NET based RAT featuring Hidden Virtual Network Computing (HVNC), encrypted communications, credential theft, and broad remote administration capabilities. Investigators also identified active and historical C2 infrastructure embedded within BNB Smart Chain transactions dating back to May 2025 and linked the threat actor to multiple phishing domains and earlier fake Docker Desktop campaigns. The FakeAgent operation highlights a growing convergence of cybercriminal tactics: abuse of trusted AI platforms through malvertising, blockchain backed resilient C2 infrastructure, and GPU assisted anti analysis techniques that enable commodity malware to bypass many traditional security tools. The campaign underscores the increasing need for enterprises to verify software sources, strengthen endpoint protections, and monitor emerging attack methods targeting trusted developer and AI ecosystems.