Key facts
- System reconnaissance, file management, on-demand extra payloads, microphone and camera capture, screen streaming, and remote desktop control are all supported by the trojan.
- Custom versions of OpenAI's ChatGPT promoted in sponsored Google results send users to malicious sites using ClickFix attacks to deliver malware.
- The page showed a fake Cloudflare check and told visitors to run a PowerShell command that starts the infection chain.
- Host reconnaissance, file searches, audio and camera capture, remote desktop access, and the capacity to launch further payloads are all provided by the RAT payload.
Blackpoint Cyber researchers said on Nov. 20, 2026 that threat actors use the ROMULUS ClickFix platform — ClickFix is a technique that tricks users into pasting commands into PowerShell or Terminal — to spread an undocumented remote access trojan, or RAT, resembling CNCMachineRMS, sending victims to malicious sites where ClickFix lures drop the RAT via DLL sideloading. Blackpoint said ROMULUS probably works as a delivery platform, pushing CNCMachineRMS and Legion Loader through separate chains while keeping ClickFix and signed binary sideloading techniques in play.
One ROMULUS payload is an MSI installer called IconEdit2Turb.msi that msiexec.exe starts after a PowerShell command runs. The installer pretends to be the Stardock Smart DeElevation Tool and begins a DLL sideloading chain that executes a malicious DLL named I++u.dll, which decodes embedded shellcode to launch CNCMachineRMS. Researchers said this RAT is the identical final implant that the ChatGPT Custom GPT lures also spread, and it contacts attacker-controlled systems through DNS over HTTPS.
According to Blackpoint, the trojan lets attackers reconnoiter systems, manage files, fetch extra payloads on demand, capture microphone and camera input, stream the screen, and take remote control of the desktop. Jonathan Semon, Huntress principal security operations analyst, said the implant is intended to gather information, observe whoever sits at the keyboard, and open a lasting path into the wider network. Semon said the typical follow-on move is to install more malware, including payloads that harvest browser data and map the endpoint.
Forescout Research first publicly documented ROMULUS last month as a malware delivery service using ClickFix-style lures to spread families such as Legion Loader, also called Satacom downloader, CurlyGate, and RobotDropper. As Blackpoint Cyber researchers Sam Decker, Nevan Beal, and Andi Ursry put it, “ROMULUS likely operates as a delivery service, distributing CNCMachineRMS and Legion Loader through separate chains”.
In phase six of the infection chain, attackers built a custom encrypted file system to hide the persistence script and the RAT. Huntress researchers described it as working like a homemade encrypted zip file: “Instead of one encrypted blob, it’s a custom archive with its own folder tree, basically a homemade, encrypted zip file”.
It starts with a small header and an index of 1,128 entries noting each item’s parent folder, size, and per-file key, followed by the file contents stored in sequence.
Separately, Huntress, a managed detection and response company, identified a ClickFix campaign in which malicious custom GPTs for OpenAI’s ChatGPT appeared in sponsored Google results and sent users to malicious sites, affecting dozens of users. The actor gave the malicious GPT the name Plus 5.6 and directed it to an alleged backup site hosted on Google Sites. Visitors saw a fake Cloudflare check and were told to run a PowerShell command that begins the infection chain; Huntress looked into at least 40 incidents connected to the Google Sites page, but confirmed that only two involved a custom GPT variant. Huntress researchers Mark O’Halloran and Jonathan Semon published their findings on Sept. 30, 2026.
The actor abused ChatGPT’s legitimate feature for building a customized version that combines instructions, extra knowledge, and skills. Malicious instructions sat on the legitimate ChatGPT.com domain, giving the operation legitimacy and making victims more likely to obey. OpenAI removed the first GPT by Sept. 25, and a second GPT tied to the campaign turned up on Sept. 27, remaining active at publication time; OpenAI plans to retire custom GPTs on Dec. 11.
ClickFix is troublesome because it tricks users into starting the compromise, through a bogus error message or CAPTCHA challenge urging them to fix an urgent device, account, or software problem. A typical attack displays obscured text and instructs the user to copy it, paste it into Windows Run, PowerShell, or macOS Terminal, and press Enter. All it takes is a hacked website, a bogus CAPTCHA overlay, and one terminal command. The technique relies on human behavior and trust in familiar tools rather than unpatched vulnerabilities, and trusted domains such as ChatGPT and Google are used to gain a foothold in this campaign.
ClickFix is growing since it bypasses email filters, web-browsing controls, and browser security features aimed at known threats. Kevin Beaumont, an independent researcher, said Reddit was filling with reports of people infected through ClickFix, describing it as “Reddit is becoming post after post after post of people getting their computer infected via ClickFix”.
Kremlin-backed hacking groups have also adopted ClickFix. Huntress reported last week that ClickFix now goes beyond simple scripts and loaders, adding persistence, lateral movement, reconnaissance, credential theft, and steganographic tricks to its chains.
Huntress researchers said most of the infection chain runs in memory or uses files that look benign, so defenders can detect based on process activity monitoring. Indicators include a PowerShell process reaching out to msiexec.exe to silently start an MSI installer from the temp folder, and a signed application running from an unusual subfolder of the Programs directory under %LOCALAPPDATA%, with a matching Run value and a scheduled task that recreates itself if deleted. TechTarget said CISOs should build layered defenses beginning with awareness training and EDR, then add e-mail filtering, web filtering, and managed browsers; defenders should watch the RunMRU registry key for suspicious PowerShell use or downloaded files, and Group Policy can restrict Run dialog access.
Semon said awareness training should shift from checking the source of something to checking what it asks the user to do. He also said no website, chatbot, support page, or verification tool should ever ask someone to paste a command into PowerShell or a Terminal to prove their identity.
In other campaigns, attackers pair ClickFix with EtherHiding on more than 70,000 compromised websites, where JavaScript queries a Polygon smart contract and then a fake Cloudflare CAPTCHA drops the slaiclind malware. Fake Zillow apartment-review pages show a fake verification prompt that installs the NetSupport RAT on Windows, and a spoofed Amazon login from sponsored Google results used an embedded ClickFix command and the Windows finger utility to deliver a Lumma Stealer variant plus NetSupport Client RMM.
In one campaign, Google Ads led users to an Anthropic Claude shared chat carrying a fake Download Claude page that tricked Mac users into running a malicious Terminal command, which deployed the Atomic macOS Stealer. A typosquatted domain, veriffication-redirect.com, served a fake Cloudflare verification prompt using clipboard manipulation to get users to run a command in the Windows Run dialog that fetched Python payloads over DNS TXT records, including an unreported trojan called CrocoRat.
In another campaign, attackers phoned targets and talked them into opening compromised WordPress sites that served ClickFix lures dropping MSI packages disguised as 3D PDF Maker Smart, Bitwarden VPN, or ESET Sysinspector, all executed through DLL sideloading. TerminalFix is a ClickFix variant that tells people to open Windows Terminal and run a command that deploys a Python tunneling implant; researchers track the activity as STAC4924.
Ads for claude mac in Google search led users through a Bing result and a compromised retail website to a fake Claude download page that used InstallFix, a ClickFix variant, to get Mac users to copy and run a malicious Terminal command; the technique is codenamed Adception. Bogus pages also imitated U.K. retailer Howdens at updates.howclens.com, DMarket, and a Czech-language Microsoft Office sign-in page at login.microsoftofice.cz with fake Cloudflare verification prompts; payloads included CrashHandler RAT, and one abused the Stripe API as its command-and-control channel.

Join the discussion