ProBackend
Vulnerabilities & Exploits
5 days ago4 min read

WordPress CVE-2026-87902: Exploitation Advances to Shell-Executing Files

WordPress CVE-2026-87902: Exploitation Advances to Shell-Executing Files

Reporting on CVE-2026-87902 describes a shift from reconnaissance to files written to disk that run shell commands when accessed. Here is what the evidence shows and what WordPress administrators should do.

AI cybersecurity threats: what this WordPress attack shows

The important change in the CVE-2026-87902 activity is concrete: requests that first appeared to be reconnaissance progressed to attempts to write files, including files that execute a shell command when opened. That is a meaningful escalation, not proof of a broad campaign or an AI-driven attack. The cited reporting does not attribute the activity to AI. BleepingComputer

Patchstack told BleepingComputer that it saw the first malicious requests at 17:44 UTC on September 22, 2026, from a small set of IP addresses targeting multiple sites under its protection. The initial traffic was described as reconnaissance. Researchers then reported a tenfold increase in related traffic and a stage involving writing files to disk. The distinction matters: probing can help identify susceptible sites; a file that carries an executable command presents a more direct risk of code execution if accessed. BleepingComputer

How CVE-2026-87902 can lead to code execution

WordPress researcher Robert Ressl discovered the issue, described as an unauthenticated path-traversal vulnerability that can lead to remote code execution under specific conditions. WordPress assigned it critical severity, with a score of 9.2 out of 10. The advisory says an unauthenticated attacker can influence get_page_template() so template resolution includes a chosen, readable local PHP file outside the active theme directories. BleepingComputer

This is conditional, not an assertion that every WordPress installation can be exploited in the same way. The reporting says the active parent or child theme must have a top-level directory beginning with page-—page-templates is one example—and the attacker must target a local PHP file that exists and is readable by the web-server account. The advisory cites pearcmd.php as an example when PHP’s register_argc_argv setting is active. It also identifies the official PHP Docker image and default cPanel configuration using PHP versions before 8.5 as affected configurations. BleepingComputer

WordPress released version 7.1.2 to address the flaw, and fixes were backported to branches down to 4.7; versions before 4.6 were not slated to receive a fix, according to the report. Administrators should check the official advisory and their actual branch rather than infer patch status from the version range alone. BleepingComputer

From reconnaissance to files on disk

Patchstack’s reported sequence is the key evidence. Early requests attempted to include ordinary WordPress core files, apparently as a way to identify vulnerable sites. Later traffic used a write-to-disk stage. Patchstack described a third stage that changed config-show to config-create, allowing pearcmd to write attacker-controlled content to a requested location. BleepingComputer says the researchers did not publish a working request example, so the observed behavior should not be turned into a speculative exploit recipe. BleepingComputer

Not every file written in the reporting was equivalent. Some payloads wrote only a string marking a host as exploitable by CVE-2026-87902. Those markers indicate testing or identification; by themselves they are not the same as a command-running payload. Separately, researchers saw payloads writing a short PHP tag that executes a shell command when accessed. That second behavior is evidence of malicious activity and raises the stakes: a web-accessible executable file can turn a vulnerability probe into a practical command-execution attempt. BleepingComputer

Reported destinations included /tmp and /var/tmp. File names included wp-pear-rce-flag.php, poc87902.php, and names in the forms luci_<random>.php and zeta_<random>.php. These are indicators reported in this activity, not a complete inventory; absence of those names does not establish that a site is clean. The report also says probes used double-encoded traversal sequences in pagename alongside a valid page_id, but it did not provide a working request example. Treat that as a detection clue, not a complete signature. BleepingComputer

What administrators should do now

Install the fixed WordPress release applicable to the site. The reporting identifies 7.1.2 as the release addressing the issue and says fixes were backported to supported branches down to 4.7. If the site is on a version before 4.6, the report says no fix was planned for that flaw; administrators should consult the WordPress advisory for the current supported path rather than assume an upgrade within that old branch resolves it. BleepingComputer

Review web-server and security telemetry for the reported indicators: requests involving the pagename and page_id pattern, and unexpected PHP files in /tmp or /var/tmp, especially the names listed above. The source specifically identifies IP addresses 169.58.48.193, 169.58.48.195, and 2001:df1:e8c0::106b for blocking. Use these as narrowly scoped indicators from the report, not as a substitute for patching or broader investigation. BleepingComputer

If a suspicious file is found, distinguish a harmless exploit marker from content that runs a command when accessed; preserve the file and relevant logs for investigation, and prevent access while responding. The source supports checking for files and blocking the listed addresses, but does not provide a full incident-remediation checklist. Avoid treating a marker alone as proof of command execution—or treating a lack of known filenames as proof that no exploitation occurred. BleepingComputer

The defensible takeaway is narrow but urgent: CVE-2026-87902 has moved beyond reported reconnaissance into file-writing attempts, with some observed files designed to run shell commands when accessed. Patch promptly, check the specific file and request indicators, and keep the evidence distinctions intact. Nothing in the cited reporting links this activity to AI or agentic systems. BleepingComputer