Particle.news

Attackers Use ChatGPT Custom GPTs to Deliver RAT Through Eight‑Stage ClickFix Chain

The campaign used sponsored Google ads and trusted chatgpt.com pages to trick users into pasting a PowerShell command that launched a multi‑stage loader and a remote‑access trojan, showing that takedowns alone cannot stop the threat

Overview

  • Huntress researchers found threat actors publishing malicious Custom GPTs on chatgpt.com that redirected users to a Google Sites page posing as a Cloudflare CAPTCHA and instructing victims to paste a PowerShell command.
  • That single command began an eight‑stage ClickFix chain that downloaded an MSI which sideloaded a patched, legitimately signed host binary (initially Canon, later Stardock) to load a concealed loader and deploy a full‑featured RAT.
  • The RAT supports remote desktop, webcam and microphone capture, file search and transfer, host reconnaissance, and the ability to fetch and run follow‑on payloads from an encrypted custom archive.
  • Huntress tied the Google Sites infrastructure to at least 40 incidents and confirmed two infections that started with a Custom GPT; OpenAI removed the first malicious GPT on Friday, September 25, and a replacement appeared on Sunday, September 27.
  • Defenders are urged to use behaviour‑based detection and the published IOCs—look for PowerShell launching msiexec on a GUID‑named MSI in %TEMP%, signed apps running from unusual %LOCALAPPDATA% folders, and reappearing Run or scheduled‑task entries—because actors rotate signed hosts and change delivery methods.