Published on 11 November, 2026 by Lana Miro
When your WordPress security scan looks clean, but the files tell another story
All responses in this series were submitted by real people and reviewed by the Melapress team. Some details may be generalized to protect anonymity.
I learned early not to trust a green light on a dashboard.
Ivica Delić runs Studio Zona, a small web agency in Split, Croatia, maintaining around sixty WordPress websites for clients. Before starting his agency, he spent 21 years working in banking, including 13 years in the payment card industry, an experience that taught him not to rely entirely on reassuring security reports.
In July 2026, this habit helped him uncover evidence of an earlier compromise on a dental clinic’s website. Eight suspicious PHP files had apparently survived a previous cleanup.
Although the files no longer contained malicious code, Ivica couldn’t determine how someone had gained access to the website or whether the security issue that allowed it had been fixed.
A security alert led to an unexpected discovery
The investigation began on a different client’s website, belonging to a restaurant.
The site’s security tool had blocked bots requesting URLs associated with gambling spam. The site had been hacked before Ivica took over its maintenance, so he investigated whether the spam was still present.
I checked the spam URLs in a browser and as Googlebot, and both returned 404, with no cloaking and no redirect. Then I searched the database on every site in the group, across all post types and statuses, and found zero spam posts. The sitemaps were clean and a site: search in Google returned nothing. So the alert was real, but the site wasn’t infected anymore.
The requests appeared to be leftovers from an earlier infection rather than evidence of an active one.
Still, the alert prompted Ivica to investigate seven other websites in the same group. All eight had a history of compromise and shared the same hosting server and IP address.
That’s when he found something more concerning on the dental clinic’s website.
Eight PHP files that should not have existed
Ivica ran a WordPress core checksum verification using WP-CLI. The result showed that every expected core file matched the official WordPress version:
mismatched: []But another part of the report caught his attention.
But the output had a second field,
should_not_exist, and it listed eight PHP files inwp-adminandwp-includes
Their names looked similar to legitimate WordPress core files, with small additions that made them easy to overlook.
All eight files returned HTTP 200 responses. But when Ivica downloaded and inspected them, he discovered they were just six or seven bytes long, containing only a PHP opening tag and no malicious code.
He suspected that an earlier malware cleanup had removed the files’ contents without deleting the files themselves. He had seen something similar on another client website hosted on the same server.
While the files were harmless at that point, Ivica was more concerned about what their presence suggested.
At some point someone was able to write a PHP file into wp-admin on that site. Nobody ever found out how they got in, and as far as I could tell, that hole could still be open. Honestly, that worried me more than the spam.
Investigating more than 12,000 files, and finding 204 false alarms
Ivica continued investigating, but the hosting provider’s firewall blocked his usual management tools, preventing him from running WP-CLI through the connector.
Instead, the client created a limited FTPS account, allowing him to scan all 12,228 files on the website.
The scan flagged 204 potentially suspicious files. After reviewing them, Ivica discovered that every alert was a false positive.
His custom scanning pattern had been too broad, flagging legitimate code in libraries such as Font Awesome, phpseclib, and Symfony Console, as well as functionality used by two code snippet plugins.
I was annoyed at myself, but I learned something. 204 flags and not one real problem. If a tool raises alarms on everything, I stop paying attention after a few weeks, and then I think I’m protected when I’m not.
By the end of the investigation, the eight empty PHP files were gone, and Ivica independently verified their removal.
Further checks found no additional evidence of active compromise. The WordPress core files matched their expected checksums, all four administrator accounts belonged to legitimate users, and checks of scheduled tasks, uploads, and database content revealed nothing suspicious. VirusTotal also reported the domain and checked the URL as clean.
For Ivica, the investigation meant additional work checking thousands of files, reviewing false alarms, and still not knowing whether the original security weakness had been fixed.
What changed after the investigation
The experience changed how Ivica checks security across his agency’s client websites.
Previously, his reports focused on whether WordPress core files matched their expected checksums. Now he also checks for unexpected files in core directories.
He also stopped treating a scanner’s confirmation that a file had been cleaned as sufficient proof that the cleanup was complete.
When a scanner says it cleaned a file, I ask what exactly it did and then I go and check the file myself.
Ivica also began checking the entire group of websites sharing the same hosting environment whenever one of them showed signs of a security issue.
He developed his own scheduled security scripts using Claude Code to check for core and plugin file changes, unexpected executable code, gambling spam, and suspicious files associated with fake CAPTCHA attacks.
Each script includes a known test case that it must successfully detect. Otherwise, the scan is treated as a failure rather than a clean result.
If a run can’t prove that the control fired, the script reports a tool failure, not a clean site.
The lesson
When a scanner finds nothing, the report alone doesn’t tell you whether the site is clean or the scanner never checked properly.
For Ivica, this experience exposed two problems with automated security checks.
The first was a reassuring checksum result that could have been mistaken for proof that everything was fine, despite eight unexpected PHP files remaining on the server.
The second was a scan that produced 204 false alarms, requiring additional investigation without identifying a single real issue.
Neither result told the full story. For Ivica, the experience changed what it means to trust a security scan. He now checks what cleanup tools have actually done and makes sure his monitoring scripts can detect the problems they’re designed to find.
The only way I can know is to make the scanner prove that it actually looked.
Have you dealt with a WordPress security incident?
Your story can show what security incidents really look like beyond the statistics, including the impact and lessons that are easy to miss.
Tell us what happened, what the experience was like, and what you learned from it. Stories may be published anonymously.