
If you click Update on a WordPress site that isn't ready for it, the update itself can be the moment someone takes the site over. That's CVE-2026-119977, and CISA put it in the Known Exploited Vulnerabilities (KEV) catalogue on 3 November 2026.
By the end of this post you'll know why the updater is exposed, what the three small mistakes behind it are, and the one server rule to apply before your next core update. It's written for anyone who runs WordPress, or tests it in a lab.
What this is, and who's affected
When WordPress updates itself, it downloads the new release and unpacks it into a working folder under wp-content/upgrade/. Once the unpack is finished, it copies those files over the live site. For a few seconds, a full copy of the new WordPress sits inside your web root.
The bug is a race condition: an attacker has to hit that working folder inside the short window between unpack and copy. Race conditions that depend on two steps happening in the wrong order are called TOCTOU bugs (time-of-check to time-of-use). Researcher Umut Bayram of Picus Security found this one and wrote it up.
Wordfence's CVE record scores it CVSS 8.1, vector CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H. Read that as: over the network, no login, no click from a user, full impact. The AC:H is the race. Ostorlab's write-up notes that WordPress itself rates the underlying race condition as Low, so you'll see the severity argued both ways.
Affected versions are 4.7.0 through 7.1.3, per Ostorlab's table. The fix shipped in WordPress 7.1.4 on 28 October 2026, with backports such as 7.0.8, 6.9.11 and down to 4.7.39. Picus adds that releases from 2.7 to 4.6 are affected too but got no fix, so those sites need to move to a supported version.
On exploitation, CISA's KEV listing is the confirmation. Ostorlab reports that Wordfence blocked 62,969 attacks aimed at it in 24 hours by 30 October, and that a honeypot operator recorded three attempts the same day. Ostorlab also says no successful compromise and no public proof of concept have been reported. I couldn't load CISA's own alert page when writing this, so the KEV date and the 6 November federal deadline come from the press and vendor write-ups.
Three small mistakes that add up to takeover
No single flaw here is dramatic. According to Picus, three ordinary behaviours line up, and any one of them fixed would have broken the chain.
The updater extracts the new release into wp-content/upgrade/, a folder that is served over HTTP and runs PHP.
The package includes setup-config.php. It finds no wp-config.php beside it and starts a fresh setup.
Setup writes a wp-config.php into the staging folder, pointing at a database the attacker controls.
The copy step doesn't skip wp-config.php, so the rogue file replaces the real one.
Mistake one: the staged tree is a second install. Because wp-content lives under the document root and PHP runs there, the unpacked wordpress/ folder can be requested like any other page while the update runs.
Mistake two: the installer can't see the real site. wp-admin/setup-config.php only checks a couple of nearby places for an existing config. It never looks up to the live site's wp-config.php. Release packages ship wp-config-sample.php, not a real config, so from the installer's view the site is brand new.
Mistake three: the copy loop trusts the staging folder. The function that copies the staged files over the live site, update_core(), builds a skip list. wp-config.php wasn't on it, so an extra config file in staging got copied across.
Once the live config points at the attacker's empty database, the site shows the install screen. The attacker creates an administrator account. On single-site installs that let admins upload plugins, that's PHP execution on the server. That last step is what turns an 8.1 into a full compromise.
Why the folder name matters
The attacker also has to know where the staging folder is. This is the part that changed over the years, and it explains the version spread.
Picus's breakdown: in 2.7.0 and 2.7.1 the path was fixed as wp-content/upgrade/core. From 2.8.0 to 4.3.35 it came from the package filename with no random part. From 4.4.0 to 5.8.18 a six-character random suffix was added, so those versions are only exploitable if the path leaks some other way, such as a verbose error. From 5.9.0 to 7.1.3 the name comes from the Content-Disposition filename, which makes it predictable again.
The attacker also needs a package type that carries the installer files. Picus says full, no_content and new_bundled packages do. Ostorlab adds that the race lasts seconds. That's short for a human and plenty for a script that keeps asking.
What 7.1.4 changed, and the catch
Per Picus, the fix has three parts, backported to every maintained branch down to 4.7:
- A 16-character random suffix on the staging folder, so the name can't be guessed.
upgrade/index.phpis no longer deleted, so directory listing can't show the random name.wp-config.phpis now on the skip list inupdate_core(), so a staged config can't overwrite the live one.
Now the catch, and it's the reason this post has a "before you update" section. Both write-ups make the same point. The unpack runs in the old updater already on your site, and only the copy runs the new code. So the update to 7.1.4 still extracts into a predictable folder. The skip list protects the copy, but Picus says an attacker could still reach the temporary installer and run PHP with web server rights before cleanup. Updates after 7.1.4 use the random path.
The practical rule: block the folder first, update second.
Block the folder before you update
Both Picus and Ostorlab give the same interim control. Deny web access to wp-content/upgrade/ at the server, then confirm it returns 403. Nginx:
location ^~ /wp-content/upgrade/ { deny all; }
Apache, in the main config or in the .htaccess in your WordPress root:
<Directory "/var/www/html/wp-content/upgrade">
Require all denied
</Directory>
Then run these checks. They're a sensible order to work in, but I haven't run them against a vulnerable install, so treat the outputs as what you should expect.
Step 4 is the detection. Picus suggests rules for requests to setup-config.php under the upgrade path, and Ostorlab says to check access logs for the same thing. There's no reason for anyone to request that file from there on a healthy site. If you can't block the folder, both write-ups offer a manual route: unpack a patched release outside the web root, copy it over without touching wp-config.php or wp-content, then run wp core update-db.
Common mistakes
- Dropping an
.htaccessinsidewp-content/upgrade/. Picus warns the old updater deletes it. Put the rule at the server or WordPress root instead. - Treating
wp core updateas a safe shortcut. Picus says WP-CLI uses the same extraction path. Ostorlab says it's fine after the folder is blocked. Both agree on the order: block first. - Removing the block after the update. The restored
index.phponly stops directory listings. Ostorlab says to keep the server block in place. - Forgetting leftovers. Delete old files under
wp-content/upgrade/and look for requests tosetup-config.phpthere. - Leaving 2.7 to 4.6 alone. Those branches got no patch. Upgrading is the only fix.
Quick recap
- CVE-2026-119977 is a race in the WordPress core updater, CVSS 8.1, in CISA KEV since 3 November 2026.
- The staged update sits in a web-reachable folder, the bundled installer thinks the site is new, and the copy step lets a rogue
wp-config.phpthrough. - 7.1.4 (28 October) fixes it, with backports down to 4.7. The update to 7.1.4 itself still uses the old, predictable path.
- Block
wp-content/upgrade/at the web server, check for a 403, then update, and leave the block in place. - No public proof of concept and no confirmed successful compromise had been reported by the sources I read.
Check this today
Run wp core version on every WordPress site you look after, then request /wp-content/upgrade/ from outside and see what comes back. Anything but a 403 means the block comes before the next update. It takes a few minutes per site.
If you test WordPress updaters in your own lab, or you've seen what these requests look like in your logs, I'd be glad to compare notes on LinkedIn.
What I left out on purpose: the exact request patterns and any working exploit steps. Neither source has published a proof of concept, and this post doesn't add one. The Picus write-up also credits an AI model as the researcher's partner and gives a four-hour timeline, which is the author's own account and which I haven't checked.
References
- Picus Security (Umut Bayram): CVE-2026-119977, when updating WordPress becomes the vulnerability
- Ostorlab: WordPress core upgrade-window RCE, fixed in 7.1.4
- WordPress.org: Version 7.1.4 release notes
- Hackread: CISA warns hackers are exploiting a WordPress core vulnerability
- CISA Known Exploited Vulnerabilities catalogue
FAQ
What is CVE-2026-119977?
A race condition in the WordPress core updater, scored CVSS 8.1 by Wordfence. During an update, the new release sits in a web-reachable folder under wp-content/upgrade/, and an attacker who reaches it can make the updater copy a rogue wp-config.php over the live one. Fixed in WordPress 7.1.4 and backported to 4.7.
Is CVE-2026-119977 exploited in the wild?
Yes, according to CISA, which added it to the Known Exploited Vulnerabilities catalogue on 3 November 2026. Ostorlab reports that Wordfence blocked 62,969 attack attempts in 24 hours by 30 October. No successful compromise and no public proof of concept have been reported so far.
Does updating to 7.1.4 fully protect me?
Not on its own. The update to 7.1.4 still runs the old extraction code, so the staging folder is predictable during that one update. Block web access to wp-content/upgrade/ at the web server before you update, and keep the block afterwards.
Which WordPress versions are affected?
Ostorlab lists 4.7.0 through 7.1.3 across every branch, with fixes in 7.1.4, 7.0.8, 6.9.11 and backports down to 4.7.39. Picus says 2.7 to 4.6 are affected too but got no fix, so those sites need a move to a supported version.
Related reading
- WordPress page template inclusion (a different WordPress core bug, in template loading rather than the updater)
- Nextcloud Collabora command injection (another web-reachable component that was never meant to be)
- Browse the whole Threats & Exploits track