Skip to main content
All editions

A 40-source toolkit went hunting for WordPress on our infrastructure

In the week of 2 November, about 40 sources ran one shared list hunting for WordPress against our infrastructure, as WordPress probes reached 73% of all sources. Requests for configuration backups, Spring Boot actuator endpoints and Apache's server-info page also climbed.

291 Intercept is our weekly report on what automated scanners went after on our own infrastructure, so you can check your systems for the same things. Each edition covers a week that ended at least seven days earlier.

A coordinated toolkit of about 40 sources went hunting for WordPress on our internet-facing infrastructure in the week of 2 November. Each ran the same list of roughly 10 paths, from servers in about 10 countries, with start times spread over several days.

It came in a week when WordPress dominated what scanners asked us for. About 590 sources requested WordPress logins, XML-RPC and plugin paths: 73% of the roughly 810 sources we saw, against an average share of 62% over the previous three weeks. Their number rose 44%, well ahead of the 26% rise in sources overall. Several other targets took a smaller slice than the week before, as the chart below shows.

Targets whose share changed from the week before
This weekThe week before
WordPress logins and files
73%
66%
Exposed source code (Git, SVN)
21%
29%
Secret and configuration files
28%
33%
Cloud key files
7%
12%

Share of all separate sources that went after each target, in percent. A source can go after several targets, so the shares add up to more than 100%.

Targets whose share changed from the week before
This weekThe week before
WordPress logins and files73%66%
Exposed source code (Git, SVN)21%29%
Secret and configuration files28%33%
Cloud key files7%12%

Forty sources, one list

The toolkit's list is a fingerprinting routine. It asks for one small file under many folder names, such as /2019/wp-includes/wlwmanifest.xml, /shop/wp-includes/wlwmanifest.xml and /cms/wp-includes/wlwmanifest.xml, and adds /xmlrpc.php. Like 82% of all sources this week, it dressed its requests in strings posing as ordinary browsers.

Separately, requests for the admin area at /wp-admin/ came from about 70 sources, 2.5 times the previous average.

How scanners find a WordPress site, then try its passwords

The file the toolkit kept asking for belongs to WordPress itself and sits at a fixed place inside an install, so asking for it is a cheap test: if it is there, WordPress probably is too. People often install WordPress in a subfolder rather than at the top of a site, so a scanner tries the same file under many likely folder names, such as a year, a shop or a CMS. Only the sites where it turns up get more of the scanner's attention.

That attention can turn to /xmlrpc.php, WordPress's XML-RPC interface. It accepts a username and password, and a method called system.multicall lets one request carry many password guesses. Wordfence has documented brute-force campaigns built on exactly this, and recommends restricting or limiting XML-RPC authentication.

One scanner, many guesses
1A scanner asks for the same WordPress file under many folder names
  • /blog/wp-includes/wlwmanifest.xml
  • /wordpress/wp-includes/wlwmanifest.xml
  • /wp/wp-includes/wlwmanifest.xml
  • /site/wp-includes/wlwmanifest.xml
  • /2019/wp-includes/wlwmanifest.xml
2Any answer tells it whether, and where, the site runs WordPress
3Then XML-RPC lets it test many passwords in a single request/xmlrpc.php

More interest in backups and server internals

Our edition for the week of 26 October noted more requests for backups, debug pages and application internals. Backups and internals rose again this week; debug pages held steady at about 120 sources.

Sources hunting for backup, dump and editor copies of files rose 42%, to about 90. Their paths included copies of WordPress's own configuration file, such as /wp-config.php.bak, /wp-config.php~ and /wp-config.php.swp. A forgotten copy like these can hand over a site's database credentials.

Application internals drew about 100 sources, up 40%. About 70 asked for /actuator/env, 2.3 times the previous average, and about 20 for /actuator/health, 6.2 times. These Spring Boot management endpoints can reveal sensitive data, according to the official documentation. In current versions only the health endpoint is exposed over HTTP by default and environment values are sanitized, so the risk lies in deployments that widen exposure or turn sanitizing off.

The steepest climb was /server-info: about 30 sources, nine times the previous average. On an Apache server with mod_info enabled and unrestricted, that page shows anyone the loaded modules and their configuration; Tenable rates it medium severity. About 50 sources asked for /storage/logs/laravel.log, 4.1 times the previous average. That file is the subject of CVE-2024-29291, which is disputed as a deployment mistake: the storage folder should never be reachable when the web root points at the public folder. If it is, the log can leak stack traces and connection strings.

A second toolkit went after secrets

A smaller group of about 20 sources shared one list of about 10 paths mixing secret files, Git configuration and application internals, including /.env, /.env.production, /.git/config, /config.json and /actuator/env. Most started within 3 hours of each other, from servers in a few countries.

Unit 42 has documented where that kind of sweep can lead: an extortion campaign retrieved environment files from about 110,000 domains, used leaked AWS keys to get into cloud accounts, then copied and deleted stored data and left ransom notes.

This was the first week we reported scanners to the networks they run on, and the ones we reported had gone after secret files.

Across all scanners, 60% of sources ran on cloud and hosting networks, with DigitalOcean and Google Cloud at 15% each. Most were drive-by visitors: 85% appeared on a single day. Activity was flat round the clock, the mark of automation.

What went quiet

Several environment-file paths, including subfolder paths of the kind we saw rising in the week of 19 October, fell back: /common/.env, /agents/.env, /.env.mail and /.next/.env had each averaged about 20 sources a week, and each fell by 60% or more. Cloud key files fell 26%, and source-code repositories fell 11% after more than doubling that week. It is too early to say whether scanners have moved on: two new Git variants, /.git/config/ and /.git/config.orig, each appeared from about 20 sources.

What to do

  • If you run WordPress and don't use XML-RPC, switch off or restrict /xmlrpc.php. Use strong, unique passwords and two-factor sign-in for /wp-admin/, and keep plugins updated.
  • Search your web root for stray copies such as /wp-config.php.bak or /wp-config.php.swp. Delete them, and change the database password if one was reachable.
  • Keep Spring Boot actuator exposure at its default and check that /actuator/env is not reachable from outside.
  • Disable Apache mod_info, or limit /server-info to trusted hosts.
  • Point Laravel's web root at the public folder so /storage/logs/laravel.log can never be fetched.
  • Keep environment files and Git metadata out of the web root, and test /.env and /.git/config from outside.

What we reported

In the week of 2 November, 291 Heat Seeker One identified about 80 servers that probed our sites for secret and configuration files, and we reported each one to the network it runs on: 15 providers in 8 countries, including Google Cloud, Microsoft Azure and DigitalOcean. Since we began reporting on 3 November, we have sent 81 abuse reports.

About these observations

This report is produced by 291 Heat Seeker One, 291 Group's security monitoring system, running on 291 Crossband, our proprietary security operations platform. It covers only traffic to 291 Group's own internet-facing infrastructure.

Figures are rounded, shares and changes are measured by the number of separate sources rather than requests, and only patterns seen from at least 20 separate sources are included. Countries are where the sources' servers are, not necessarily where the people running them are. Kinds of network come from who operates each network; only large, well-known providers are ever named.