WordPress RCE attack chains 2025/2026: web shells, crypto mining bots and SEO cloaking – analysis and hardening for admins
From the flaw to the compromised installation: wp2shell and co., web shell and backdoor persistence, crypto mining bots and SEO cloaking – with locations, indicators and a hardening checklist for admins.

A WordPress takeover rarely ends with the first executed PHP file. Attackers secure further access, place malicious code in several locations and then use the server for mining, spam or redirects. If you only delete the visible web shell, you can soon be facing the same infection again.
This article is aimed at admins who need to understand such attack chains and investigate an affected installation. The CVE overview classifies the entry points; afterwards it covers persistence, detection and cleanup. For an introduction without technical details there is the general article.
The essentials at a glance
- The core attack chain “wp2shell” (CVE-2026-63030 + CVE-2026-60137), published in July 2026, affects 6.9.0 to 6.9.4 and 7.0.0 to 7.0.1. The patches appeared on 17 July 2026 in 6.9.5 and 7.0.2; the critical part of the chain has a CVSS score of 9.8. Wiz reported successful attacks and large-scale scanning on 20 July 2026. WordPress GHSA · Wiz
- Plugins and themes were also affected by critical flaws in 2025 and 2026, some with observed exploitation (W3 Total Cache, WordPress File Upload, Everest Forms, Simple File List, Drag&Drop File Upload for CF7, Ads Pro unpatched).
- Patchstack recorded 11,334 new vulnerabilities in the WordPress ecosystem in 2025, 42 percent more than in 2024. The total covers different severity levels. Patchstack: State of WordPress Security 2026
- After the entry point, web shells and additional access via must-use plugins, manipulated configuration files or cron are common. Mining, SEO spam and redirects are possible consequences; the StopAndProtect and DollyWay campaigns show the abuse as criminal infrastructure.
1. The CVE landscape: core and top plugins
1.1 WordPress core: wp2shell (CVE-2026-63030 + CVE-2026-60137)
In wp2shell two flaws interplay: a confusion of REST batch routes bypasses access checks, and an SQL injection allows further steps up to code execution. Remote Code Execution (RCE) here means that attackers can run their own code on the server.
- CVE-2026-63030 (CVSS 9.8): The REST batch endpoint (
/wp-json/batch/v1or?rest_route=/batch/v1) desynchronises validation and handler dispatch with a malformed sub-request → auth bypass on protected endpoints. - CVE-2026-60137 (CVSS 5.9): SQL injection in the
author__not_inparameter ofWP_Query(a scalar instead of an array is inserted unchecked intoNOT IN). In the chain: UNION-based database extraction → admin account → plugin upload/web shell. - Indicators:
POST /?rest_route=/batch/v1with HTTP 207/200, user agentwp2shell/rezwp2shell, web shell directorieswp-content/plugins/wp2shell_<hex>/, PHP inwp-content/cache/, temporary files under/tmp/. Elastic
The research for 2025 contains no comparable core RCE chain. The examined security release 6.8.3 addressed an authenticated information leak and XSS. No general statement about the absence of bugs in the core can be derived from this.
1.2 Critical plugin and theme vulnerabilities
The following selection contains direct RCE flaws and other paths to a site takeover. The required user privileges are given in the last column. The patch versions are minimum versions for the respective flaw, not a recommendation to stay on that version.
| CVE | Product | CVSS | Affected | Patched in | Type |
|---|---|---|---|---|---|
| CVE-2026-27384 | W3 Total Cache | 9.8 | ≤ 2.9.1 | 2.9.2 | Unauth. arbitrary code execution; follow-up CVE-2026-5032: security token derivable from user-agent → mfunc tags with PHP execution |
| CVE-2025-9501 | W3 Total Cache | 9.0/8.1* | < 2.8.13 | 2.8.13 | Unauth. command injection via _parse_dynamic_mfunc |
| CVE-2024-11613 | WordPress File Upload | 9.8 | ≤ 4.24.15 | 4.25.0 | Unauth. RCE + file read/delete (family: CVE-2024-9939, CVE-2024-11635) |
| CVE-2025-5394 | Theme “Alone” | 9.8 | ≤ 7.8.3 | 7.8.5 | Unauth. file upload → RCE; exploited from 12.07.2025, >120,000 attacks blocked |
| CVE-2025-1128 | Everest Forms | 9.8 | ≤ 3.0.9.4 | 3.0.9.5 | Unauth. upload/read/delete → RCE (via .csv→.php rename) |
| CVE-2025-34085 | Simple File List | 10.0 | < 4.2.3 | 4.2.3 | Unauth. unrestricted file upload (Metasploit module) |
| CVE-2025-3515 | Drag&Drop File Upload for CF7 | 9.8 | ≤ 1.3.8.9 | 1.3.9.0 | Unauth. web shell upload (.phar, .php5, .inc missing from blacklist) |
| CVE-2025-4689 | Ads Pro | 9.8 | ≤ 4.89 | unpatched per research | Unauth. LFI → RCE |
| CVE-2025-0366 | Jupiter X Core | 8.8 | ≤ 4.8.7 | 4.8.8 | Contributor+ SVG upload → RCE |
| CVE-2024-9849 | Real 3D FlipBook | 8.8 | ≤ 4.8 | 4.8.5 | Author+ arbitrary file upload |
| CVE-2026-4350/4351 | Perfmatters | 8.1/8.1 | ≤ 2.5.9.1 | 2.6.0 | Subscriber+ path traversal → file delete (wp-config.php → full site takeover) |
* CVE-2025-9501: WPScan/INCIBE 9.0; Wordfence considers exploitation “theoretical” (8.1) – the different assessment should be borne in mind when comparing figures.
1.3 Takeover through bypassed access checks
Not every entry point leads directly to code execution. An unauthorised admin login or a disclosed secret can however enable the next step. These cases should therefore be considered separately from direct RCE flaws:
- CVE-2023-28121 WooCommerce Payments (CVSS 9.8, ≤ 5.6.1): unauth. privilege escalation; peak of 1.3 million attacks on 157,000 sites in a single day (14.07.2023). SecurityWeek
- CVE-2023-32243 Essential Addons for Elementor (9.8, >1 million installs): password reset without validating the reset key; millions of probing attempts after the patch.
- CVE-2024-10924 Really Simple Security (9.8, >4 million sites): 2FA auth bypass; “one of the most severe vulnerabilities in their 12-year history” (Wordfence). Wordfence
- CVE-2022-31474 BackupBuddy (7.5): unauth. arbitrary file download; ~4.9 million attacks blocked, zero-day from 26.08.2022.
- InfiniteWP Client (≤1.9.4.4, no CVE): unauth. login as admin; public Metasploit module.
For Contact Form 7, WPForms, Yoast SEO, Rank Math, Jetpack, WPBakery and Slider Revolution, the accompanying research found no critical RCE flaw in the period considered. Recorded were XSS, file reads and information leaks, among others. For Slider Revolution, CVE-2025-9217 and CVE-2025-10249 concerned reading files with Contributor privileges or higher. That is a research finding, not a security clearance for these products.
2. From entry point to persistent access
A web shell enables further actions on the server. Additional persistence ensures that access survives the deletion of individual files. Both layers are relevant for an investigation.
2.1 Entry and persistence

- Unauth. RCE/upload (see above) or auth-bypass chain (SQLi/auth, then admin → plugin installation). Observed with wp2shell:
POST /wp-admin/update.php?action=upload-pluginwith a malicious plugin upload or directlySQL INTO OUTFILEtowp-content/cache/. Wiz - Web shell types: one-liners
<?php eval($_POST['...']??'http_response_code(404);');as well as disguised 150 kB web shells as “plugin” (gzip+base64, runtime unpacking viaeval(gzuncompress(base64_decode(...))), own REST route withpermission_callback => '__return_true', passthru handler). Wiz - Backdoor persistence:
eval(base64_decode(...))one-liners at the start ofwp-config.php/ themeheader.php/functions.php. Wordfence 05/2026- mu-plugins backdoors (
wp-content/mu-plugins/*): loaded automatically and separately visible in the backend under “Must-Use”; cannot be deactivated there, only by removing the file from the load path. WordPress documentation Examples 2025:wp-index.php(ROT13 payload,wp_optionsstorage, admin user “officialwp”),redirect.php,custom-js-loader.php. Sucuri 07/2025 · Sucuri 03/2025 - Current example 2026 (StopAndProtect): an installer writes
wp-content/mu-plugins/wp-sec.php, which provides a hidden REST endpointwp-sec/v1/uploadwith hardcoded credentials and explicitly allows.phpuploads (RCE potential) – afterwards the installer deactivates and deletes itself. Check Point Research 08/2026 - Cron persistence: WP-Cron events + hosting cron jobs (rebuild after cleanup). Sucuri 06/2019
- Outside the web root: backdoors in the directory tree next to webmail/phpMyAdmin;
auto_prepend_filein.user.ini/.htaccess. Wordfence 05/2026 - Self-replication: mu-plugin samples copy themselves to
mu-plugins/andplugins/, create admin accounts (wp_optionsoptionbu), own string decoder instead ofeval. Wordfence 09/2026
2.2 Where to look
The following locations help narrow down an investigation. Individual file names, status codes or requests are not proof of compromise on their own. Content, origin and the interaction with further traces are decisive.
- Files:
mu-plugins/*,db.php,advanced-cache.php,.user.iniwithauto_prepend_file, PHP underwp-content/uploads/andwp-content/cache/, unusually long one-liners inwp-config.php, cache/snippet files, hidden admin users (optionbu,officialwp, …), file manager web shells such aswp-cssv.php/wp-verifyup.php(StopAndProtect). Check Point 08/2026 - Access logs:
POST /?rest_route=/batch/v1, UAwp2shell/rezwp2shell, HTTP 207 on batch endpoint,admin-ajax.phpLFI attempts (template=../../../wp-config),plugin-install.php?tab=upload. - Scheduled tasks and processes: audit hosting cron + WP-Cron events; check web server and system logs for hints of
/tmpuploads and unusual processes. Not every process start is recorded there automatically. - Service workers in the admin area, if there are hints of browser-side persistence. The linked support article is a single case report, not general evidence for this reinfection path. WP Support Forum
2.3 Cleanup and rebuild

Before deleting suspicious files, the affected installation should be isolated and the findings preserved. Afterwards the goal is to close the entry point and remove all remaining access:
- Patch the exploited flaw in core, plugin or theme. Without this step, the next infection can follow immediately.
- Reinstall core, plugins and themes from trusted original sources.
- Also search outside the web root, e.g. in neighbouring webmail directories, logs and backup files.
- Check the database:
wp_options, unwanted cron events, foreign admin accounts and manipulated options belong to it. - Renew credentials and enable 2FA. Remove malicious hosting cron jobs and clean UP WP-Cron; legitimate tasks must be kept.
3. Crypto mining bots: indicators and defence
With server mining, the miner runs as a process on the host. For the investigation, process data and network connections are therefore just as important as the files in the WordPress directory. The Group-IB report describes a Linux campaign; the following Sucuri case documents its use on a WordPress site. - Group-IB, campaign since May 2026: heavily disguised XMRig 6.25.0 botnet implant (source banner “PRIVATE VERSION FOR BOTNET”); self-deleting (runs only in memory, mutex /tmp/.lock), process masquerading (e.g. as ssh), XOR-encrypted configuration, mining pool domain unable[.]download, campaign ID “V25 (Generation 26)”; persistence via redundant cron jobs in unmonitored user accounts (PAM pam_rootok abuse as forensic smokescreen). Group-IB, 07/2026 - WordPress case (Sucuri 2022): a dropper downloads xmrig-6.17.0-linux-x64.tar.gz (hidden e.g. in /1/1.html), unpacks it, mines Monero; a restart loop and cron jobs (* * * * *, i.e. every minute) reload miner and configuration as needed. Sucuri 2022 - Indicators: miners under /tmp, /var/tmp; process masquerading (bind mount, names such as irqbalance, systemd-logind, dbus-daemon); connections to mining pools (pool.minexmr.com, c3pool, unable[.]download) or high TCP ports; disabled logging services (rsyslog/auditd). AhnLab 2023 · AhnLab 07/2026 · Group-IB - MITRE ATT&CK T1496.001: XMRig used e.g. by Blue Mockingbird, TeamTNT, Skidmap, Kinsing, LoudMiner, Lucifer. MITRE
To be distinguished from this is browser mining via JavaScript, e.g. the historical Coinhive. The cases described here concern server mining. Unusual, heavily obfuscated <script> blocks in the page source should still be investigated.
For detection and defence: - Hosting monitoring on CPU/processes; block mining pool domains and known IOCs; cron control. - Store logs externally and as tamper-proof as possible; alert when rsyslog or auditd stops. In case of suspicion also examine memory: a miner that deletes its file after starting is no longer visible in a file scan. - If malicious cron jobs or backdoors remain active, the miner can return after a cleanup. Sucuri 02/2023
4. SEO cloaking and SEO spam: detection and cleanup
Google calls the delivery of different content to users and search engines for search manipulation cloaking. Attackers use this separation to hide spam from site owners. Google Spam Policies · Google web.dev - Delivery by user agent (Googlebot/Bingbot/Baiduspider), referrer (google.com) and/or mobile UA. - Current case (Sucuri, 01/2026): malware at the end of a legitimate PHP file (pages/index/IndexHandler.php) checks the user agent, loads spam via cURL from the attacker server (fugthatshit[.]baby/desktop/kindipublisher.txt, IP 104.21.50.131), outputs the spam for bots and ends the application with exit;. Normal visitors still see the original page. The report explains the mechanism using a PHP application; it does not prove a WordPress-specific entry flaw. Sucuri 01/2026 - DollyWay / VexTrio (GoDaddy, 02/2025): >20,000 compromised WP sites in eight years, >10,000 active sites with ~10 million impressions/month in early 2025; distributed TDS network on hijacked websites; redirect chains to scams/malware/apps (AdsTerra/PropellerAds monetisation); reinfection on almost every page view, WPCode snippets as persistence track. Dark Reading/GoDaddy 03/2025
Locations: wp-content/uploads/ (spam generators such as cache/spam.php), wp-content/uploads/.spam/, foreign sitemap files (spamurl.txt in the root registered as Search Console sitemap), DB custom HTML widgets/“Simple Custom CSS and JS”/WPCode, hidden admin clones, PHP files at the end of manipulated app/plugin files. Segurium/Sucuri 08/2026 · Sucuri SiteCheck H1/2024 (PDF)
Other documented campaigns (context): - Chinese gambling spam campaign: >50,000 sites, title switching, Baidu Automatic Push abuse. Sucuri 09/2022 - Japanese keyword hack: generator PHP in directories where PHP must not run; after cleanup check the security messages in the Search Console. Segurium - Gootloader: compromised WP sites as fake forums (Q&A/Legal Templates); IP blocklists against admins; delivers REvil/Cobalt Strike/Kronos. Sophos 2025
Detection: - Compare two requests: curl -A "Mozilla/5.0 (compatible; Googlebot/2.1; +http://www.google.com/bot.html)" <url> and a request with a normal user agent. Unexpected spam content is suspicious; different answers alone do not prove cloaking. A clean test does not rule it out either, because delivery can also depend on IP address, referrer or cookies. - site: search for spam keywords (viagra, casino, ‘Situs Toto’, Japanese terms); the cache: operator has been deactivated since 02/2024 → Search Console/URL inspection instead of cache view. - .htaccess audit: unexpected RewriteRule blocks, ErrorDocument hijack, auto_prepend_file, patterns index.php?smsite=…. - Audit WPCode/“Simple Custom CSS and JS” endpoints and DB widgets (DollyWay persistence).
During cleanup, the redirect raises the root-cause question: which access or which malicious code produces it? Files, database, cron and admin accounts must be examined together. Afterwards work through the security issues in the Search Console and, if offered, request a review. SiteGuarding 11/2026
5. Why attacks reach many sites quickly
Automated scanners look for known endpoints and vulnerable extensions. A single scan in the log is not proof of a successful break-in. The following metrics and campaigns show the scale; their periods and counting methods differ and should not be added together.
- Scan/attack targets:
wp-login.php/XML-RPC (brute force;system.multicallas amplifier),admin-ajax.php, REST (/wp-json/wp/v2/users,?rest_route=/batch/v1),plugin-install.php?tab=upload, known plugin paths. Wordfence XML-RPC - Wordfence 2024 figures: >54 billion malicious requests blocked, >55 billion password attacks from ~136 million attack IPs; XSS attempts 9 billion, SQLi 1.1 billion. Wordfence 2024 report
- StopAndProtect (2026, Check Point): thousands of WP sites as infrastructure (malware hosting, C&C, log/loot storage); ~2,000 compromised domains in the logs, 6,000+ affected IPs; internal project names
0a_botnet/fake-captcha; one scanned target ran a 2021 WordPress version with ~40 vulnerabilities. Check Point 08/2026 - DollyWay / VexTrio (GoDaddy 2025): >20,000 WP sites over eight years as redirect/TDS network; competing malware removed, own reinfected on page load. Dark Reading/GoDaddy 03/2025
- Balada Injector (campaigns until 2024): ~1 million compromised sites estimated; CVE-2023-3169 → >17,000 sites; CVE-2023-6000 Popup Builder → 6,700+/3,900+ sites; rogue admin users + arbitrary plugin installation. Sucuri 04/2023 · BleepingComputer 10/2023 · THN 03/2024
- Sign1 (since 08/2023): 39,000+ sites in 6 months (~57,000 by mid 2024); XOR-encoded JS, URLs change every 10 minutes, only fires on Google/Facebook/Instagram referrers; TDS → VexTrio. BleepingComputer 03/2024
- DNS TXT record TDS (2024): 23,820 sites in H1/2024; redirect URLs from DNS TXT answers of attacker domains. Sucuri 04/2024
- SocGholish/FakeUpdates: ~11.88% of injections in H1/2024; fake browser update landing pages on hacked sites. Check Point 03/2024
6. What admins can set up preventively
The measures complement each other. Updates close known flaws, restricted rights limit the damage, and monitoring helps to detect a break-in earlier.
Sources (selection)
- WordPress GitHub Advisory GHSA-ff9f-jf42-662q (wp2shell)
- Wiz Research: wp2shell · Elastic Security Labs: wp2shell Detection
- Check Point Research: StopAndProtect (08/2026)
- Group-IB: XMRig Covert Ops (07/2026)
- Sucuri: Cloaked SEO Spam Attack (01/2026)
- Dark Reading/GoDaddy: VexTrio/DollyWay (03/2025)
- Patchstack: State of WordPress Security 2026
- Wordfence 2024 Annual WordPress Security Report
- Google Search Central: Spam Policies · Google web.dev: Fix the cloaked keywords hack
- MITRE ATT&CK T1496.001
- AhnLab ASEC: Mining indicators Linux
- Further CVE references are documented in the accompanying research notes.
Editorial status: 7 November 2026. All CVE data and campaign figures are backed by primary sources in the accompanying research notes. An accessible introduction without technical details: the general article.