Attackers stage a RAT lure on the real chatgpt.com using a Custom GPT
Huntress traced a RAT campaign that hides its first-stage lure in a ChatGPT Custom GPT on the real chatgpt.com, across at least 40 incidents. Because the page sits on a trusted domain, blocking the ClickFix PowerShell paste is the control that works.
The Watch · Security desk
What happened
- Victims arrive through a sponsored Google result for "ChatGPT" that opens a Custom GPT titled "Plus 5.6," dressed up to look like a new ChatGPT model.
- The GPT tells users the service is limited and to continue on a "backup domain," a Google Sites page with a fake Cloudflare CAPTCHA that has them paste a PowerShell one-liner into Windows Run.
- OpenAI removed the first reported GPT on September 25, and a second GPT tied to the same campaign was already online two days later.
- Microsoft Defender quarantined the malicious MSI in at least one case, but by then the malware had already written a Run key and a scheduled task.
- A later wave swapped Canon's signed executable for a Stardock-signed one, and Huntress confirmed two Custom GPT-driven infections in the set.
Why it matters
- Who's exposed exposure Domain reputation and URL allowlists cannot flag a page served from chatgpt.com, so a trusted-domain check offers no protection against this lure.
- What it limits constraint The PowerShell reaches its download server by a decimal IP address that Windows resolves but some URL filters do not parse, narrowing what URL filtering catches in the download step.
- Who has to decide decision Every variant runs through the same user action, pasting PowerShell into Windows Run, and that is the step defenders can block regardless of the hosting domain or the code-signing.
Next story · Security · 5h ago
X attributes more than 13,000 suspended accounts to Iranian, Russian and Chinese influence operationsConfirmed2 publishers
Run the pasted command and the campaign moves through eight stages, each one hiding the next [9]. The one-liner fetches a 27,000-character script written as negative numbers; the script decodes itself in memory, so nothing readable touches the disk [11]. It then installs ISOSimple.msi, which poses as an "Advanced Printer Configuration Reader" from a publisher called Softplicity and hides from the list of installed programs [12].
The core of the payload is DLL sideloading through a legitimately signed Canon executable, COTFileReadApp.exe from Canon's CaptureOnTouch product [6]. Huntress described it plainly: "When a program asks for a DLL by name, Windows checks the program's own folder first, so whatever sits next to the EXE with the right name gets loaded. That's DLL sideloading, and it's the door the attackers used." [7] The attackers left Canon's genuine, signed binary alone and swapped its logging library for a copy that still exports the functions Canon expects but also imports one extra DLL, rdCore.dll, where the malicious code sits; the only difference is a single line in the import table [8].
rdCore.dll pulls an encrypted loader out of a WAV file bundled with the installer. The file opens as real audio and turns to payload bytes partway through, and its header claims about 700KB while the file actually runs past 1MB [16]. More than 300KB of it is the hidden payload [18]. A rolling XOR cipher decodes the loader in memory [17], which then opens monitor.raw, a custom encrypted archive of 315 folders, 806 files, a persistence script, and the final remote access trojan, each file under its own key [13].
Security Affairs described that archive as a homemade encrypted filesystem built to move the trojan past automated tooling [15].
What to watch
- Whether OpenAI adds checks that stop Custom GPTs from redirecting users to external "backup" domains.
- Whether the campaign rotates to further signed executables after the Canon and Stardock waves.
- Whether the same ClickFix chain shows up seeded from other trusted platforms.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence70
- Adoption20
- Hype gap+30
- Incentives40
- Confidence68
Perspective Coverage
8 publishers- Builder
- Builder 30%
- Operator
- Operator 64%
- Investor
- Investor 6%
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Huntress researchers found attackers abusing ChatGPT's Custom GPT feature as a malware delivery mechanism across at least 40 related incidents.
- [2]
Huntress said the payload establishes resilient access through dual persistence and DLL sideloading, using a Canon-signed executable and, in a later wave, a Stardock-signed one, and that it confirmed two Custom GPT-driven infections among the incidents.
- [3]
A victim searches Google for "ChatGPT," clicks a sponsored result, and lands on a Custom GPT page at the real chatgpt.com domain titled "Plus 5.6," designed to look like a new ChatGPT model.
- [4]
The GPT replies that the service has limited availability on its primary domain and to continue on a "backup domain," a Google Sites page serving a fake Cloudflare CAPTCHA; from there a ClickFix attack has the victim paste a PowerShell one-liner into Windows Run that downloads and installs a malicious MSI.
- [5]
Huntress reported the first Custom GPT to OpenAI, which took it down on September 25; two days later another GPT linked to the same campaign was already online.
- [6]
The technical core of the campaign is DLL sideloading through a legitimately signed Canon application, COTFileReadApp.exe, from Canon's CaptureOnTouch product.
- [7]
When a program asks for a DLL by name, Windows checks the program's own folder first, so whatever sits next to the EXE with the right name gets loaded. That's DLL sideloading, and it's the door the attackers used.
- [8]
Canon's binary is genuine and its signature checks out; the attackers replaced its logging library with a modified version that still exports the functions Canon expects but also imports one extra DLL, rdCore.dll, where the malicious code lives, changing only one line in the import table.
- [9]
The attack uses eight stages, with each stage hiding the next.
- [10]
The PowerShell command uses a decimal IP address instead of the usual dotted format, which Windows accepts but some URL filters may not recognize.
- [11]
The PowerShell downloads a 27,000-character script made up of negative numbers that decodes itself in memory, so the readable code is never saved to disk.
- [12]
It installs an MSI called ISOSimple.msi that pretends to be an "Advanced Printer Configuration Reader" from a publisher called Softplicity and hides itself from the list of installed programs.
- [13]
The loader opens monitor.raw, a custom encrypted archive containing 315 folders, 806 files, a persistence script, and the final remote access trojan, each file under its own encryption key.
- [14]
Microsoft Defender detected and quarantined the MSI in at least one case, but by then the malware had already created a Run key and a scheduled task.
- [15]
Security Affairs described the monitor.raw archive as a homemade encrypted filesystem built to package and deliver the RAT past automated tooling.
- [16]
rdCore.dll extracts an encrypted loader from a WAV file bundled with the installer; the file begins with real audio and switches to encrypted payload partway through, and its header claims about 700KB while the file is actually over 1MB.
- [17]
A rolling XOR cipher decodes the loader directly into memory.
- [18]
The WAV file's declared size understates it by more than 300KB, the space that holds the hidden payload.
Sources
8 independent publishers whose own reporting we read for this story.
- bleepingcomputer.comCustom ChatGPTs push ClickFix attacks to deploy RAT malware
1 article · September 29, 2026
- helpnetsecurity.comMalicious Custom GPT on chatgpt.com lures users into installing a RAT
1 article · September 29, 2026
- huntress.comAttackers Abuse ChatGPT Custom GPTs to Deliver RAT via ClickFix
1 article · September 28, 2026
- infosecurity-magazine.comAttackers Combine ChatGPT Feature Abuse With ClickFix to Deliver Trojan Malware
1 article · September 30, 2026
- scworld.comMalicious ChatGPT Custom GPTs lead to ClickFix attack
1 article · September 30, 2026
- securityaffairs.comAttackers Abuse ChatGPT Custom GPTs to Deploy a Full-Featured RAT
1 article · September 30, 2026
- securityweek.comHackers Use ChatGPT Custom GPTs in ClickFix Attacks
1 article · September 29, 2026
- thehackernews.comAttackers Abuse ChatGPT Custom GPTs to Deliver RAT via ClickFix Lures
1 article · September 30, 2026
Topics and entities
Topics
- AI Platform Abuse and Account Fraud
- DLL sideloading and signed-binary abuse
- ClickFix
- Malvertising