Cybersecurity researchers at Huntress have uncovered a malicious campaign abusing custom ChatGPT models and Google Sites to infect Windows devices with malware. Attackers set up custom bots hosted on the legitimate ChatGPT platform to trick users into executing malicious code disguised as system verification steps.
Custom GPTs Used as Entry Points
The attack begins with custom GPTs configured on ChatGPT.com, giving the initial link an appearance of legitimacy. In one instance detailed by Huntress, attackers named a custom bot "Plus 5.6" to imitate official model releases from OpenAI. When users interacted with the bot, it displayed a message alleging limited availability on the primary domain and directed victims to a backup link.
Exploiting Google Sites and ClickFix Tactics
The redirect leads users to a webpage hosted on Google Sites. Once on the page, visitors encounter a fake Cloudflare CAPTCHA prompt designed to execute a ClickFix social engineering attack. The prompt instructs users to copy and execute a snippet of code directly into the Windows Run prompt to resolve the fake verification issue.
Payload Capability and Operational Scope
Executing the command deploys a Remote Access Trojan (RAT) identified by Huntress as "@input". The malware grants operators remote access to infected Windows devices, allowing them to monitor screens, activate webcams and microphones, search local files, assess installed security software, and retrieve secondary malware payloads via encrypted communication channels.
According to Huntress, its security operations center addressed at least 40 incidents connected to the campaign's Google Sites domain. OpenAI took down the initial custom GPT on September 25, 2026, following notifications from researchers, though a similar custom bot emerged two days later.
AI-generated editorial illustration; not a documentary photograph.
Source: Hackers host fake ChatGPT model on its official website — but really it's just malware
SIGNAL TITANS
Leave a Reply