A single request to the WordPress REST API can trigger an authorization bypass, facilitate progressive database reading, and execute code on the server. Though patched since July, this exploit chain known as WP2Shell is still being targeted, as observed by CrowdSec, with increasing frequency.
The WP2Shell exploit chain, patched on July 17, serves as a reminder that some of the most insidious WordPress vulnerabilities do not stem from plugins. This vulnerability affects the core content management system and combines two flaws that, when considered independently, do not convey the full extent of the attack. The first flaw reroutes a nested request to an incorrect access manager, while the second flaw converts a query parameter into a channel for database reading.
In response to this exploit chain, WordPress released patches 7.0.2 and 6.9.5 on July 17, 2026, just days before the CISA added CVE-2026-63030 to its catalog of exploited vulnerabilities on July 21. The company enforced automatic updates for the affected versions. Telemetry released by CrowdSec indicates that their network continues to log attempts targeting this exploit chain, though it remains unclear why the targeted sites are still reachable.
The Nested Request Uses the Wrong Access Control
The batch processing function of the REST API represents the first technical link in this chain. A call to /?rest_route=/batch/v1 can encompass multiple sub-requests, including one directed at the /wp/v2/categories route. The vulnerability labeled CVE-2026-63030 lies in how WordPress determines the handler used to check authorization for these nested requests. A routing confusion causes it to apply a check that is not appropriate for the intended route.
The routing confusion of CVE-2026-63030 does not instantly grant the attacker admin console access. However, it does allow the attacker to access code paths that should be protected without authentication. This is where the second vulnerability, CVE-2026-60137, comes into play. The author__not_in parameter of WP_Query can accept a value that reaches the SQL query construction process without sufficient sanitization. As a result, the authorization bypass opens the door, while SQL injection provides the means to query the database.
Response Time Becomes a Reading Channel
WP2Shell employs a so-called time-based injection. The response does not directly provide the requested records. Instead, the attacker submits a condition and asks the server to delay its response using a SLEEP function when the condition is met. A slower response indicates a “yes,” while a faster response means “no.” By repeating the query character by character, an automated system can reconstruct usernames, password hashes, and other sensitive information stored in the database.
The time-based injection effectively transforms its own delay into an exfiltration channel, responding one response at a time. An automated extraction replaces the visible download of a table with a long series of responses, the durations of which carry information. The secrets obtained can then be used to impersonate a privileged account or alter site configurations. WordPress confirms that the combination of these two CVEs can even lead to remote code execution. This chain progresses from a routing error to total server control, all without legitimate user interaction.
Sixty-Two Thousand Addresses Don’t Equate to That Many Attackers
CrowdSec reports observing 62,802 unique IP addresses over a span of fifteen days, with 3,147,518 signals detected over thirty-four days. According to their telemetry, the daily volume has remained relatively stable, averaging around 92,574 signals. Meanwhile, the number of distinct sources grew from approximately 1,900 per day during the first week to 10,486 by August 21. This variation suggests a broader distribution of exploitation efforts towards an increased number of infrastructures, rather than simply an escalation by a small group.
However, these statistics only reflect the scope monitored by CrowdSec, a firm that also sells application protection. They are not audited by a third party and do not represent 62,802 individual attackers or compromised sites. A single operation might leverage a network of proxies, infected machines, or recycled addresses. A signal corresponds to a detected attempt, not to a successful execution. The key takeaway remains the expansion of monitored sources, rather than a global count of attackers or victims.
Update to WordPress 7.0.2 or 6.9.5
The listing of CVE-2026-63030 in the CISA’s KEV catalog on July 21 serves as a critical priority marker. The NVD has indeed acknowledged its active exploitation. Therefore, for any organization, running a vulnerable version and having indicators of exploitation takes precedence over the severity score attached to the vulnerability.
This urgency calls for an immediate update to WordPress 7.0.2 or 6.9.5, which eliminates the flaws in the code. WordPress maintains that earlier branches before 6.9 are not affected by this specific exploit chain, but that does not mean they should be treated as secure without examining their support status. Implementing an application firewall rule or a virtual patch can block this known query form during deployment. While this reduces exposure, it neither fixes the routing confusion nor the SQL injection in the instance.
Beyond this patch, accessible WordPress sites form the perimeter to inventory, including staging environments, migration clones, and any abandoned installations. It is essential to verify the executed version, not merely rely on a reported successful update task. For instances exposed prior to the patch, reviewing queries directed to the relevant REST routes, database errors, and account changes should complement the deployment of the patch based on the logs retained. A corrected machine is not necessarily a sanitized one if the attacker has already extracted its secrets.



