Executive Summary
Between July 21 and July 22, 2026, a malvertising campaign dubbed 'FakeAgent' targeted at least 29 organizations by leveraging legitimate Claude.ai infrastructure. Attackers used SEO poisoning to direct users to a public Claude Artifact on the official domain, which then redirected victims to a spoofed download page for a 'Claude Desktop' application. The delivered payload was SectopRAT, a remote access trojan designed to exfiltrate sensitive user data, credentials, and financial information.
The campaign is technically sophisticated, employing 'EtherHiding'—the practice of hosting command-and-control (C2) configuration within Ethereum blockchain transactions—to ensure infrastructure resilience. Additionally, the malware utilized advanced anti-analysis techniques, including VMProtect, GPU-based anti-VM gates that check graphics hardware strings (e.g., QEMU, VMWare), and a custom DirectX shader-based decryption routine to obfuscate its primary .NET payload. This approach significantly complicates traditional sandbox and static analysis efforts.
This threat highlights the increasing risk of adversaries abusing trusted AI platforms to host malicious content. Organizations should be aware that even links on legitimate top-level domains like claude[.]ai can host unverified, user-generated malicious artifacts.