Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB
CVE-2026-6712 Overview
The Website LLMs.txt plugin for WordPress contains a Stored Cross-Site Scripting (XSS) vulnerability [CWE-79] affecting all versions up to and including 8.2.6. The flaw resides in the plugin's admin settings handling, where input sanitization and output escaping are insufficient. Authenticated users with administrator-level permissions can inject arbitrary web scripts into pages, which execute when any user accesses the affected page. The vulnerability only affects WordPress multi-site installations and installations where unfiltered_html has been disabled.
Critical Impact
Authenticated administrators on multi-site WordPress deployments can inject persistent JavaScript payloads that execute in the browsers of any visiting user, enabling session theft and content manipulation.
Affected Products
- Website LLMs.txt plugin for WordPress, all versions through 8.2.6
- WordPress multi-site installations using the plugin
- WordPress installations where the unfiltered_html capability has been disabled
Discovery Timeline
- 2026-04-21 - CVE-2026-6712 published to NVD
- 2026-04-22 - Last updated in NVD database
Technical Details for CVE-2026-6712
Vulnerability Analysis
The Website LLMs.txt plugin exposes administrative settings that accept user-controlled input intended for storage and later display in WordPress pages. The plugin fails to apply adequate sanitization to incoming data and fails to escape values during output rendering. As a result, an authenticated administrator can supply HTML or JavaScript payloads through the settings interface. These payloads persist in the WordPress database and execute in the browser of any user who loads a page containing the injected content.
Stored XSS in plugin-rendered output is particularly impactful because the payload triggers without further attacker interaction. On WordPress multi-site networks, super-administrators control privilege boundaries, and site administrators are not implicitly trusted with raw HTML. Disabling unfiltered_html is a common hardening step that elevates this defect to a security boundary violation rather than expected behavior.
Root Cause
The root cause is missing or incomplete input sanitization on plugin settings combined with absent output escaping when rendering values into HTML contexts. Functions such as sanitize_text_field(), wp_kses_post(), and output-time escaping with esc_html(), esc_attr(), or esc_url() were not applied to user-controlled values [CWE-79].
Attack Vector
An attacker must first obtain administrator-level credentials on a target WordPress site within a multi-site installation, or on a site where unfiltered_html has been revoked. The attacker then submits a crafted payload through the plugin's admin settings. The payload is stored persistently and rendered to subsequent visitors of the impacted pages. The high privilege requirement and the conditional environment lower the practical risk, which is reflected in the CVSS base score.
No verified public exploit code is available. Technical specifics are documented in the Wordfence Vulnerability Report and the WordPress Changeset Update.
Detection Methods for CVE-2026-6712
Indicators of Compromise
- Unexpected <script> tags, on* event handlers, or javascript: URIs stored in wp_options rows associated with the website-llms-txt plugin.
- Outbound browser requests from visitor sessions to attacker-controlled domains immediately after loading pages rendered by the plugin.
- Modifications to plugin settings recorded in audit logs that originate from administrator accounts outside expected change windows.
Detection Strategies
- Inspect database entries for the plugin's option keys and search for HTML tags, encoded script payloads, or escape sequences.
- Compare current plugin file hashes against the vendor-released patched version to confirm integrity.
- Review web server access logs for POST requests to /wp-admin/options.php and plugin-specific endpoints submitted by administrator accounts.
Monitoring Recommendations
- Enable WordPress audit logging to capture changes to plugin settings, including the actor account and submitted values.
- Deploy a Content Security Policy (CSP) with strict script-src directives to neutralize injected inline scripts.
- Monitor administrator account authentication events for anomalous geolocations, user agents, or session reuse patterns.
How to Mitigate CVE-2026-6712
Immediate Actions Required
- Update the Website LLMs.txt plugin to the version released after 8.2.6 that contains the referenced changeset fix.
- Audit existing plugin settings and remove any stored values containing HTML or script content.
- Rotate administrator credentials and enforce multi-factor authentication on all privileged WordPress accounts.
- Review WordPress multi-site administrator assignments and remove unnecessary privileged users.
Patch Information
The vendor addressed the issue in the changeset published on WordPress.org. Refer to the WordPress Changeset Update for the exact code changes applied to sanitize input and escape output in the plugin's admin settings handler. Site operators should apply the update through the WordPress admin Plugins interface or via WP-CLI.
Workarounds
- Temporarily deactivate the Website LLMs.txt plugin until the patched version is installed.
- Restrict administrator-level access on multi-site networks to a minimum set of trusted accounts.
- Implement a web application firewall (WAF) rule to block script tags and event handler attributes submitted to plugin settings endpoints.
# Update the plugin using WP-CLI
wp plugin update website-llms-txt
# Verify the installed version after update
wp plugin get website-llms-txt --field=version
# Temporary workaround: deactivate the plugin until patched
wp plugin deactivate website-llms-txt
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB

