Critical WordPress RCE Discovered by AI Enables Pre-Auth Admin Takeover and Server Access

A critical pre-authentication remote code execution (RCE) vulnerability has been uncovered in WordPress, marking a significant milestone in cybersecurity research because the flaw was discovered by an advanced artificial intelligence system rather than a human researcher.

The vulnerability, attributed to OpenAI’s GPT-5.6 Sol Ultra model, demonstrates how AI-assisted security research is rapidly evolving and is capable of identifying highly complex exploit chains that may be difficult for traditional vulnerability discovery methods to detect.

Security experts believe this represents one of the most sophisticated AI-driven vulnerability discoveries reported to date.

Critical WordPress RCE Discovered by AI

The vulnerability affects WordPress’s Batch API (/wp-json/batch/v1), a feature introduced in WordPress 5.6 that allows multiple REST API requests to be bundled into a single HTTP request.

According to the technical analysis, the issue stems from the Batch API’s two-stage processing model. During the first stage, WordPress validates every request contained within the batch, while the second stage executes them.

Under specific malformed request conditions, the validation results become misaligned with the corresponding execution handlers.

This synchronization failure enables an attacker to validate one endpoint while executing another, effectively bypassing parameter validation and sanitization on batch-accessible REST endpoints.

GPT-5.6 Sol Ultra reportedly identified that this validation bypass could be combined with a previously known sanitization weakness affecting the author_exclude parameter in the GET /wp/v2/posts endpoint.

When this parameter is supplied as a scalar string instead of the expected array, WordPress skips integer casting and directly incorporates the supplied value into an SQL query.

Although the Batch API normally does not support GET requests, the AI-generated exploit chain cleverly bypassed this limitation by nesting one batch request inside another, enabling attackers to reach the vulnerable endpoint without requiring authentication.

Rather than stopping at SQL injection, the research demonstrated an advanced exploitation technique that transformed database access into complete administrative compromise.

Since WordPress stores passwords using secure hashes, simply reading database contents would not provide immediate administrator access.

Instead, the exploit manipulated WordPress’s in-memory post cache using UNION-based SQL injection to fabricate fake post objects during request processing.

By abusing the platform’s oEmbed caching mechanism and leveraging post hierarchy cycle detection logic that resets parent references, the exploit forced WordPress to reconcile attacker-controlled in-memory objects with legitimate database records.

This technique enabled the creation of a forged customize_changeset post, which WordPress uses to stage theme and configuration updates associated with an administrator account.

Once the crafted changeset was applied, the request temporarily inherited administrative privileges. The exploit then leveraged the parse_request action hook, restarting WordPress’s request parsing process while maintaining the newly elevated permissions.

This allowed a pending request to create a new administrator account to execute successfully despite initially originating from an unauthenticated user.

After obtaining administrator privileges, achieving remote code execution became straightforward. The attacker could simply authenticate using the newly created administrator account and upload a malicious plugin ZIP file through WordPress’s standard plugin installation interface.

Once installed and activated, the plugin would execute arbitrary code on the web server, resulting in complete system compromise.

To help administrators assess their exposure, Searchlight Cyber has released a free scanning tool that checks whether WordPress installations may be vulnerable before public proof-of-concept exploits become widespread.

Independent security researchers Calif and Hacktron have also reportedly reproduced the vulnerability shortly after its disclosure, further validating the findings.

Given that WordPress powers more than 500 million websites worldwide and the vulnerability requires no authentication in default MySQL-backed deployments, security professionals recommend treating this issue as a top-priority patching event.

Organizations running affected WordPress versions should immediately apply available security updates, restrict unnecessary REST API access where possible, continuously monitor administrator account creation events, and review plugin installations for unauthorized activity.

The discovery also highlights the growing role of AI in vulnerability research, signaling a future where artificial intelligence may become both a powerful defensive tool and a catalyst for uncovering increasingly sophisticated security flaws.