In a significant leap for cybersecurity, GPT-5.6 Sol Ultra, an advanced AI model, has reportedly identified a critical remote code execution (RCE) flaw in WordPress. The breakthrough was achieved using just $25 worth of AI resources, illustrating the potential of AI in vulnerability detection.
Advanced AI Models in Cybersecurity
Researchers at Searchlight Cyber employed GPT-5.6 Sol Ultra to scrutinize a local copy of WordPress’s source code. The model, with the help of four AI agents, diligently worked for over six hours to uncover this major vulnerability.
The AI unearthed an exploit chain that could let unauthorized attackers obtain administrative access and execute arbitrary code on susceptible WordPress sites. This discovery highlights the evolving role of AI in identifying and mitigating cybersecurity threats.
The Value of the Discovered Vulnerability
The potential impact of this vulnerability is substantial, with researchers estimating damages up to $500,000. This figure underscores the value of RCE vulnerabilities, particularly those affecting widely adopted platforms like WordPress.
Given that WordPress supports over 500 million websites worldwide, a default-configuration RCE poses a significant risk. The reported flaw centers around WordPress’s Batch API, accessible via the /wp-json/batch/v1 endpoint, which processes multiple REST requests within a single HTTP request.
Technical Details of the Exploit
The vulnerability arises from a misalignment between validation outcomes and matched REST handlers, allowing an attacker to validate one request while executing another. This desynchronization bypasses parameter sanitization safeguards within REST API routes.
Further investigation revealed an issue within the posts endpoint, where an author_exclude value could be unsafely sent to a database query. The AI model devised a nested batch request to circumvent method validation, leading to a pre-authentication SQL injection vulnerability.
Exploiting this SQL injection, the AI transitioned to RCE by manipulating WordPress’s in-memory post cache and oEmbed caching mechanism. By creating controlled database-backed post records, the AI forged changeset objects associated with the administrator user, temporarily assuming administrative privileges.
The attack culminated in using WordPress hooks to execute a malicious batch request, allowing the creation of a new administrator account and installation of a harmful plugin.
Response and Recommendations
Searchlight Cyber has postponed public disclosure to allow time for addressing this issue. Independent researchers, including Calif and Hacktron, have successfully replicated the exploit, and proof-of-concept code is now available on GitHub.
Administrators are advised to apply the latest WordPress security updates promptly and inspect logs for unusual requests to the /wp-json/batch/v1 endpoint. Organizations can assess their exposure using the wp2shell hosted scanner.
Strengthening cybersecurity measures by accelerating threat detection and integrating solutions like ANY.RUN with security operations centers is crucial for defending against such vulnerabilities.
