Manual
Product site My Page

Security / Speed

Last updated · Published · Sorabun Editorial Team

Watch daily for known vulnerabilities, security fixes, tampering and suspicious files, close entry points with two-factor authentication and more, and serve images as WebP for faster pages

The most common way in is a plugin that wasn't updated

"Sorabun > Security / Speed" watches and closes the ways a WordPress site gets taken over (Pro). The settings that make images lighter and pages faster also live here.

Most WordPress takeovers exploit weaknesses the developer has already fixed. After a fixed version is released, sites that haven't updated get targeted. But WordPress shows "update available" the same way for a new feature and for a security fix, so the urgent updates get buried.

This screen checks the following automatically every day, and notifies you when something urgent turns up:

  • Whether installed plugins, themes or core have a known vulnerability
  • Whether an available plugin update contains a security fix
  • Whether WordPress core, plugin and theme files have been modified, or suspicious files have been placed
  • Whether an unfamiliar administrator has been added, whether scripts have been planted in the database, and whether the site URL has been changed

You don't need to sign up for any paid service or API key. It contacts WordPress.org and the Sorabun license server (for matching known vulnerabilities).

1. Known vulnerabilities

Installed plugins, themes and core are matched against a public vulnerability database (Wordfence Intelligence). The Sorabun license server imports the database every day.

ShownMeaning
SeverityCritical, High, Medium, Low (from the CVSS score)
Action "Update to x.x or later"A fixed version is out. Updating to it fixes the problem
Action "No fix yet"Not fixed yet. Updating can't fix it, so if you don't use it, consider deactivating or deleting it
  • Only plugin and theme names (slugs) are sent for matching. Installed versions and site contents are not sent. Versions are compared on your own site
  • Critical and High issues, and issues with no fix yet, are notified on the day they're found
  • License activation is required
Source

Vulnerability information comes from Wordfence Intelligence (free for personal and commercial use, and redistribution is permitted under its terms). The source, plus the copyright notices and licenses of the records shown (Wordfence/Defiant, MITRE for CVEs, etc.), appear at the bottom of the screen. Only Sorabun's server holds the Wordfence API key; your site never contacts Wordfence directly.

2. Updates that contain security fixes

For each plugin with an available update, it reads the changelog from your installed version up to the latest version from WordPress.org and checks whether a security fix is mentioned. Results are listed most urgent first.

VerdictMeaning
Update nowThe changelog clearly mentions a security fix, such as "vulnerability", "XSS", "SQL injection" or "CVE-". Update as soon as you can
Looks like a safety fixThe changelog mentions fixes that likely relate to safety, such as sanitizing input or permission checks
Can't check the changesThere is no changelog, or it can't be read (themes, plugins distributed outside WordPress.org, etc.)
Feature updateNo security fix was found in the changelog

A small update within the same WordPress release line (e.g. 6.8.1 → 6.8.2) is almost always security and bug fixes, so it's shown as "Update now".

While there's an "Update now" item, a red notice appears on every admin screen. It disappears once you update.

Keep updating even without "Update now"

Security fixes the developer doesn't write in the changelog can't be detected. "Update now" means at least do this one right away. Apply the other updates regularly too.

Apply only security fixes automatically

Turn on "Automatically apply only updates that contain security fixes" in "4. Close the entry points" to have WordPress's automatic updates (twice a day) apply only "Update now" items and plugins with a fixed version for a known vulnerability.

WordPress's automatic updates are all-or-nothing, but this applies only the urgent ones. It doesn't change the automatic update settings of other plugins. Applied updates are recorded on the screen.

3. Tampering and suspicious files

WordPress core

It compares WordPress core files (wp-admin and wp-includes) against the official list of correct files (checksums) published by WordPress.org. It finds three things:

  • Modified files: malicious code has been added to a core file
  • Missing files: a core file has been deleted
  • Unknown PHP inside the core folders: a common form of the "back door" left behind after a takeover

wp-content (themes, plugins, images) isn't compared, because its contents change with legitimate updates.

If differences turn up and you don't know why, use "Re-install version x.x.x" in "Dashboard > Updates" to put the official files back (your posts and settings are not deleted). Unknown PHP files aren't removed by re-installing, so check their contents in your server's file manager and delete them.

Development versions can't be checked

Development versions of WordPress, such as betas, have no official list, so the screen says the check isn't possible. That doesn't mean tampering was found.

Plugins, themes and suspicious files

"Scan files" checks two things (it also runs automatically every day).

  1. Matching (reliable): plugins distributed on WordPress.org have an official list of correct files for each version. Installed files are compared against it to find modified files and PHP files that aren't in the original package. Themes distributed on WordPress.org are compared the same way, using that version's package downloaded from WordPress.org
  2. Pattern check (a guide): files that can't be matched, such as the uploads folder and paid or custom plugins and themes, are checked for patterns common in files planted during a takeover (running hidden code, running values sent from outside, PHP in the uploads folder, PHP disguised as an image, and so on)

Files that match the official files are not pattern-checked, so legitimate plugins and themes are rarely flagged.

About matching themes

For themes, WordPress.org doesn't publish a "list of correct files" the way it does for plugins. So the package (zip) of the installed version is downloaded from WordPress.org and compared with the files inside. Up to 3 packages are downloaded per scan, and the result is remembered for a week per version (the same version isn't downloaded again and again). Only themes that WordPress.org recognizes in WordPress's update check can be matched (paid, custom and child themes go through the pattern check). If you edit theme files yourself, those files show up as "modified".

For anything found, "Ask AI to explain" has AI describe in plain words what the code does, whether it looks malicious, and what to do. Only 25 lines around the finding are sent to AI.

LevelMeaning
HighAlmost certainly malicious (notified on the day it's found)
GuideCommon in malicious code, but legitimate code uses it too. Check the contents
Not a replacement for a professional scan

The pattern check doesn't have the coverage of the large signature sets professional security companies maintain. "Nothing found" doesn't guarantee safety, so if you suspect a takeover, consult a professional.

Administrators

A takeover often adds one administrator as a back door for getting in again later. "Check now" (and the daily watch) lists administrators created in the last 30 days and people who became administrators, and compares them with the activity log.

ComparisonMeaning
There is a record in the activity logSomeone who was logged in created the account or raised its role. The person and the date are shown
Became an administrator through an action by someone who wasn't logged inThere is a record, but nobody was logged in. This is what an exploited weakness looks like (work done with WP-CLI or on the server also shows this way)
There is no record in the activity logThe activity log was recording, but there is no record. This is an administrator added without Sorabun knowing, for example written directly into the database
Can't be checkedIt happened before the activity log started recording

"No record" and "not logged in" are notified on the day they're found. The number of administrators and their names are listed too, so check for any name you don't recognize. Remove the role from, or delete, any administrator you don't recognize on the "Users" screen (Sorabun doesn't delete them).

The first check remembers the current administrators as "people who were already there". Administrators found in later checks are shown as "recently became an administrator" for 30 days from the day they were found.

Database

Scripts that send visitors to another site are often written into the database (post content, widgets, theme settings, the settings of plugins that add code to the header, and so on) rather than into files. "Check now" (and the daily watch) looks for the following.

What it findsLevel
The site URL (home, siteurl) has been changed. Compared with the value from the previous check. If the activity log has a record of a logged-in person changing it (in "Settings > General", for example), the change is acceptedHigh
A script that runs hidden code (a combination such as eval with atob, fromCharCode, base64 or runs of \x)High
A script with embedded content (data:), or hidden code run from an attribute such as onerrorHigh
A script loaded from another domain (fine if it's an ad or analytics tag)Guide
The domains of the site URL (home) and the WordPress URL (siteurl) differGuide

To avoid calling legitimate things suspicious:

  • A <script> alone isn't called suspicious. It looks at combinations
  • Common ad, analytics and embed sources are treated separately and only counted: Google Analytics, Tag Manager, AdSense, the Meta pixel, Microsoft Clarity, X and Instagram embeds, Yahoo! tags, A8.net, Moshimo Affiliate and so on. Look-alike names (such as googletagmanager.com.example.net) aren't treated separately. Neither are hosts where anyone can put files (jsDelivr and the like), because takeovers use them
  • Structured data (application/ld+json) isn't executed, so it isn't checked
  • Revisions, the trash, temporary caches and Sorabun's own records (the firewall log contains attack strings as they are) aren't checked

If you changed the URL yourself and it still says it was changed, open the admin screen at the new URL and press "Check now". The change is then accepted.

Fixing is done by hand

Findings are only reported; Sorabun doesn't delete anything (deleting an ad tag by mistake would stop your revenue). In a post, switch the editor to "Code editor" to find and remove a script. If the site URL was changed, set it back in "Settings > General" and change everyone's password right away.

4. Close the entry points

Every setting is off by default. Some of them stop features certain sites need, so read the description before turning one on.

SettingWhat it preventsKeep in mind
Automatic security fixesApplies urgent updates without waiting for someoneAn update can occasionally break the layout. Applied updates are recorded on the screen
Require two-factor authentication for administratorsEven if a password leaks, nobody can log in without the phoneWhen on, administrators who haven't set it up can't open other screens until they do. Set up your own first
Leaked passwordsStops people from using passwords that leaked from other sites and are circulating. They're refused when a password is set, and if a login uses one, a change is recommendedChecks use Have I Been Pwned. The password itself isn't sent: only the first 5 characters of its hash are sent, and matching happens on your site
Login brute forceAfter 5 failures in a row from the same connection, login is blocked for 15 minutes. Stops attacks that keep trying passwords by machine. Attacks that try a little from many connections are met with an image CAPTCHA (explained below)If many people log in from the same network (an office, for example), one person's repeated typos can lock everyone out for a while
Username leaksHides the "?author=1" trick that reveals login IDs, and the user lists in the REST API and sitemap, from visitors who aren't logged inAuthor pages (/author/name/) still display
XML-RPCTurns off an old integration mechanism used for attacks that try many passwords at onceThe WordPress mobile app, Jetpack and some external posting tools will stop working
PHP in the uploads folderStops PHP from running in the uploads folder, against attacks that upload PHP disguised as an image and open itWorks on servers where .htaccess applies (Apache, LiteSpeed). On nginx, ask for the one line shown on the screen to be added to the server configuration

When a login is blocked, the date and time, the connection (partly masked) and the username that was tried stay on the screen. Blocking only uses connection information that can't be faked.

Attacks that try a little from many connections (per username and site-wide)

Blocking per connection can't stop an attack that tries once each from many IPs. But blocking per username would let anyone who knows a username lock the administrator out. So with "Login brute force" on, an image CAPTCHA is used instead of a lockout.

What happenedWhat it doesWhen it goes back
Logins to the same username failed from 5 or more connections within 60 minutesAsks for the image CAPTCHA for that username (no lockout; the real person can log in by typing the characters)60 minutes (extended while failures continue)
Failures across the whole site reached 30 within 10 minutesAsks everyone for the image CAPTCHABack to normal 30 minutes after the failures stop
  • The same person tried by email address and by login name counts as one person. Usernames that don't exist (such as admin) are counted too
  • Even if "Image CAPTCHA" is off on the screen, the CAPTCHA appears on the login screen only at these times (PHP's GD is required. On servers without GD, only the per-connection lockout works)
  • During that time, XML-RPC only accepts application passwords (XML-RPC can't show a CAPTCHA). The real person can log in from the login screen (with the CAPTCHA) and with a passkey
  • Attempts stopped by the CAPTCHA, attempts during a lockout and XML-RPC attempts during that time can't get in, so they don't count toward the lockout. The record on the screen also shows "Asked for the CAPTCHA for this username" and "Asked everyone for the CAPTCHA"
  • The numbers are set so that typos from home, work and a phone don't reach them
Fewer database writes during an attack

During an attack, failures and requests arrive many times a second. Writing to the database each time would slow the site down by itself. With an object cache (Redis, Memcached, etc.), failures are counted there and not written to the database; without one, each failure is written once. Attempts during a lockout or stopped by the CAPTCHA aren't written. The firewall log merges the same connection and the same attack into one entry per minute (with an object cache, it counts them and writes them together every minute).

Firewall

Stops obvious attack requests inside WordPress.

  • SQL injection (peeking into the database), reading server files, running commands, injecting PHP, cross-site scripting
  • Probes for config and hidden files such as .env, wp-config.php.bak and .git/
  • Bursts of requests for pages that don't exist (30 within 5 minutes): robots hunting for vulnerable plugins are locked out for 30 minutes
ModeWhat it does
OffThe default
Log onlyRecords without blocking. Start here for a few days and check that no legitimate traffic shows up
BlockReturns "403" for matching requests

To avoid locking out legitimate users:

  • Requests from people who can write posts (logged-in editors and administrators) aren't checked (HTML and code in post content is normal)
  • Password fields aren't checked
  • For bursts of missing pages, search engines such as Google and Bing are verified by reverse DNS and excluded (their self-reported names can be faked, so they aren't trusted)
  • AI search crawlers (OAI-SearchBot, Claude-SearchBot, PerplexityBot and similar) can't be verified by reverse DNS, so a client that identifies as one is blocked only after 120 hits instead of 30 (it is still blocked eventually, and a client that drops the AI name mid-burst is blocked on the spot)
  • IPs and ranges listed in "Allowed connections" are never blocked. Adding your office's fixed IP is a good idea

The log merges the same connection and the same attack into one entry per minute (so writing the log doesn't slow the site down during an attack).

What it can't do

It can't stop requests for files the server returns directly (existing images, files that don't go through WordPress), or floods of traffic (DDoS). Use a service such as Cloudflare alongside it if you need that.

Country blocking (login screen)

Enter two-letter country codes separated by commas (Japan is JP, the US is US) in "Countries allowed to log in", and the login screen and XML-RPC can only be opened from those countries. Admin logins are often attempted from abroad, so if your site is only used from Japan, setting just JP turns away most brute-force attempts at the door.

  • If the country can't be determined (a lookup failure, for example), access is allowed
  • It won't save if your current connection's country isn't included (so you don't lock yourself out the moment you save)
  • When logging in from a trip abroad, add your IP to "Allowed connections" or add the country
  • Countries are determined with DB-IP's list (IP Geolocation by DB-IP, CC BY 4.0), updated monthly. On sites behind Cloudflare, Cloudflare's determination is used

Sites behind Cloudflare

On sites behind Cloudflare, every request arrives from a Cloudflare IP. Left as is, everyone would look like the same person and a login lockout would lock everyone out. So the visitor IP that Cloudflare adds is used only when the request really comes from Cloudflare (the same header coming from anywhere else can be faked, so it isn't trusted).

Sites behind a CDN or load balancer other than Cloudflare (trusted proxies)

Sites behind a hosting company's load balancer or a CDN other than Cloudflare also receive every request from that device's IP, so everyone looks like the same person. The login lockout, the inquiry rate limit, the number of tries for LINE-only posts and so on would apply to everyone.

Write that device's IP or range (in a form such as 10.0.0.0/8 or 2001:db8::/32; IPv4 and IPv6 both work) one per line in the firewall's "Trusted proxies", and the visitor IP is read from the header the device adds, only for requests that come from that range.

  • Choose the header to read from X-Forwarded-For (the default), Forwarded or X-Real-IP. Pick the one your device adds
  • The header is read from the right. Each device a request passes through adds to the right of X-Forwarded-For, so the left side can be written freely by the visitor. Reading from the right, it skips the IPs of trusted proxies (and Cloudflare) and treats the first remaining IP as the visitor
  • Headers on requests from outside the range aren't read (connecting directly and writing X-Forwarded-For doesn't make you someone else)
  • Ranges that are too wide (wider than /8 for IPv4, wider than /16 for IPv6) can't be added, because anyone could then fake the visitor IP
  • If it's empty, headers aren't used, as before (the default)

Open "What this request shows" in the setting to see the current request's connection (REMOTE_ADDR), the contents of the headers, and the connection Sorabun uses. If REMOTE_ADDR is the device's IP and your own IP appears in X-Forwarded-For or similar, your site needs this setting. Check the device's IP in your hosting company's or CDN's documentation.

When saving country blocking, the "don't lock yourself out" check uses how you appear with the new "Trusted proxies" you're saving.

How do you set up two-factor authentication? (each user's profile)

Each user sets up two-factor authentication under "Two-factor authentication" in "Users > Profile".

  1. Press "Set up" to show a QR code
  2. Scan it with an authenticator app on your phone (Google Authenticator, Microsoft Authenticator, 1Password, etc.)
  3. Enter the 6 digits shown in the app and press "Verify". It turns on and shows 10 backup codes. Each can be used once if you lose your phone, so print them or save them in a password manager

From then on, you're asked for the 6 digits after your password. You can get the 6 digits wrong up to 5 times; after that you start over from the password. The same 6 digits can't be used twice.

  • The QR code is created on the screen itself (the secret key isn't sent to any outside service). The secret key is stored encrypted, and backup codes are stored in a form that can't be reversed
  • If someone loses both their phone and backup codes, another administrator can press "Remove this user's two-factor authentication" on that user's profile, and they can log in with just the password
  • For XML-RPC, users with two-factor authentication can only log in with an application password

Passkeys (log in with fingerprint, face or PIN)

Press "Add a passkey" under "Passkeys" on your profile screen to log in with the fingerprint, face or PIN of the device you're using (phone or computer). From then on, just press "Log in with a passkey" on the login screen, with no password.

  • The site only stores half of the key pair (the public key), so a leak from the site can't be used
  • The key only responds to the site it was registered for, so it won't be used even if you're lured to a fake site (resistant to phishing)
  • It checks that you have the device and your fingerprint or PIN at the same time, so you aren't asked for the two-factor 6 digits when logging in with a passkey
  • Up to 10 per user. When you change devices, delete old ones from the list
  • Passkeys work on https sites (a browser rule)
  • You can also save them in a password manager such as 1Password (Using 1Password)

Change the login URL, image CAPTCHA and login alerts

These are the same protections as SiteGuard WP Plugin. You'll find them in the "4. Close the entry points" table.

SettingWhat it preventsKeep in mind
Change the login URLMoves the login screen to https://your-site/any-name. wp-login.php, plus /wp-admin/, /login and /admin while logged out, return "Not Found", so bots hammering wp-login.php have nothing to tryUse 6 or more lowercase letters, digits, - or _. The new URL is also emailed to the person who changed it. Bookmark it. Password-protected posts, form submissions (admin-post.php) and Ajax keep working. Not available on multisite
Image CAPTCHAAsks for the 4 characters in an image on login, password reset and registration. Optionally also on comments from people who are not logged inThe image is drawn on your server, so no outside service is used (PHP GD is required). Easily confused characters (0, O, 1, I, L) are not used. Lowercase and full-width input are accepted
Login alertsEmails people who can write posts (Contributor and above) when they log in from an unfamiliar device (browser or network)Not sent for every login. The first login after turning it on only remembers that device. Up to 10 devices are remembered per person
If you lock yourself out

Add the line define( 'SORABUN_SECURITY_OFF', true ); to wp-config.php using your server's file manager. All settings in this section (two-factor authentication, login lockouts, the login URL change, the image CAPTCHA, etc.) stop. Monitoring and notifications continue. Remove the line once you can log in again. If you only forgot the login URL, define( 'SORABUN_LOGIN_URL_OFF', true ); stops just the login URL change (wp-login.php works again).

5. Lighter images for faster pages

Images are most of what slows a page down. Converting JPEG and PNG to WebP makes them a fraction of the size with almost no visible difference.

Turn on "Serve as WebP" to display article images from their WebP copy when one exists.

  • The original image isn't modified. A WebP copy is created next to it (photo.jpg.webp for photo.jpg) and swapped in only when displayed
  • Turning it off goes back to serving the original images immediately
  • Newly uploaded images get a copy right away
  • For existing images, use "Make existing images lighter in bulk". It works 15 images at a time and shows progress and how much was saved
  • Images that would get bigger as WebP (such as well-compressed PNGs) get no copy and are served as the original
  • Only images in this site's uploads folder are swapped. Images from other sites are left as they are
  • If you no longer need the copies, "Delete the WebP copies" removes them. The originals stay
Images made by AI are already saved lighter

Images Sorabun makes with AI, such as eyecatches and diagrams, are saved as WebP at a maximum width of 1600px. You can switch this with "Save AI-generated images lighter" in "Settings > Generation defaults". Slides for video stay PNG so video editing software can read them.

If the server's image processing (GD / Imagick) doesn't support WebP, copies can't be created. The screen tells you when that's the case; ask your hosting company about WebP support.