A fake Cloudflare check can install a WordPress backdoor using your own admin session

Securonix Threat Research published its analysis of an attack chain it calls TRUST#GRAFT on 22 October, and it has been circulating through the security press since. It deserves a second look from anyone managing WordPress sites for clients, because it does not break WordPress authentication at any point — it waits for an administrator who is already logged in, and uses that session against them. That is the kind of thing nobody catches by reading a changelog, and the reason a standing integrity check on the sites you manage is worth more than another scanner.

How the chain works

The loader runs as injected JavaScript on a compromised page:

  1. Fake verification overlay — a full-screen iframe mimicking a Cloudflare “Just a moment” screen covers the real page
  2. Fingerprinting and a proof-of-work gate — the script requests a token and difficulty value, brute-forces a small nonce, and submits browser fingerprint and campaign data before the server returns anything useful. This is a delivery decision point, and it filters out automated analysis
  3. Credential scraping — `localStorage` and `sessionStorage` are searched for token-like key names, with up to 1,500 characters of `document.cookie` posted to a reporting endpoint
  4. The admin-session pivot — the script checks `/wp-admin/`, pulls the plugin-upload page, extracts the WordPress nonce, downloads a staged `wm.zip`, posts it to `update.php?action=upload-plugin` as multipart form data, then fetches `plugins.php` and follows the activation link

Every one of those requests carries the administrator’s own same-origin cookies and a nonce WordPress itself issued. From the application’s point of view, an admin installed a plugin. Because that is literally what happened.

The payload hides in the one directory people forget

`wm.zip` contains a PHP plugin presenting as Web Media Optimizer, version 2.1, with a performance-oriented description. It is a backdoor. On activation it:

  • Copies itself to `wp-content/mu-plugins/wmedia-recovery.php` — must-use plugins load automatically and cannot be disabled from the Plugins screen
  • Hides itself from plugin listings, action links and update responses
  • Injects remote JavaScript into public pages, with controller config stored in WordPress options and transients
  • Disables TLS verification when fetching its controller source
  • Exposes token-gated paths that can set an administrator auth cookie or swap the remote script source

The `GRAFT` half of the name is the mu-plugins copy. Delete the visible plugin directory and the site is still backdoored. Securonix is explicit that a complete cleanup has to check `wp-content/mu-plugins` directly and confirm no equivalent copy remains.

Reported infrastructure includes `caapman[.]me`, `vtevens[.]icu`, `gerlace[.]com`, `nutmann[.]club` and `okuneva[.]sbs`, with the payload consistently at `/p/wm.zip`.

The habit it exploits is an agency habit

TRUST#GRAFT targets one specific working pattern: staying logged into wp-admin in the same browser profile used for ordinary browsing. Reviewing a client’s live site in one tab, wp-admin in another, research and links in a third. That is a normal Tuesday for most people who do this work, and it is the whole precondition for the attack.

Which makes the mitigation unusually cheap:

  • A separate browser profile for wp-admin. No extensions, no general browsing, nothing else logged in. This alone breaks the chain
  • Log out of admin sessions rather than leaving them open across days
  • Watch `wp-content/mu-plugins/` for new PHP files — any new file there is worth a question, not just `wmedia-recovery.php`
  • Audit WordPress options and transients for remote URLs that nobody put there
  • During containment, rotate admin credentials and invalidate all sessions, including application passwords
  • Where you can, require an approval step for plugin installation rather than leaving upload rights on every admin account

Why this one is awkward to detect

There is no vulnerable version number to compare against, so a plugin audit will not flag it. The upload looks like legitimate admin activity in access logs. The proof-of-work gate means scanning the compromised page from a server will often return nothing. The plugin hides from the Plugins screen. The realistic detection points are file-level: a new file in mu-plugins, an unexpected `POST` to `update.php?action=upload-plugin`, or outbound requests to `/p/wm.zip`.

That is a monitoring problem, not a patching problem — which is the uncomfortable part, because patching is the bit most agencies have covered and monitoring is the bit they do not.

What this means for Codelibry clients

If you are on one of our maintenance plans, file integrity monitoring already covers the must-use plugin directory and unexpected plugin installs, admin session and credential hygiene is part of the standing routine, and anything found gets written up in your monthly maintenance report. Nothing is needed from you on this one. If you are not on a plan, this is the category of problem the plan exists for: an attack that leaves the plugin list, the version numbers and the uptime graph all looking perfectly normal, and only shows up if somebody is actually watching the files.

Source: Securonix

Vitalii Omelchenko
Founder at Codelibry and WordPress enthusiast. Helping digital agencies to protect their margins and do better at website delivery.Need help with wordpress builds? Book a call using the Contact page
our Blog

Explore our Latest Insights

The FBI just tied over 1,000 hacked WordPress sites to $28 million in ransom payments
News
November 18, 2026
319 WordPress plugin vulnerabilities in one week. Fourteen were critical and one is still unpatched
News
October 2, 2026
The WordPress backdoor that rebuilds itself. Deleting the files is not a cleanup
News
October 2, 2026
whatsapp icon