GPT-5.6-found wp2shell WordPress RCE now exploited across 500M sites

Security researcher Adam Kues used OpenAI’s GPT-5.6 to discover wp2shell, a pre-authentication remote-code-execution chain in WordPress core that affects more than 500 million sites and is now being exploited in the wild.

Kues, a researcher at Searchlight Cyber, said the AI-assisted hunt took about 10 hours and roughly $25 of GPT-5.6 usage. “No security researcher could have found and completed this exploit chain in 10 hours without AI,” he said in Searchlight Cyber’s account of the discovery. The speed is the striking part: a frontier model compressing what would normally be weeks of manual code review into a single working day.

The chain combines two flaws that need no authentication or user interaction: CVE-2026-63030, a REST API batch-route confusion, and CVE-2026-60137, a SQL injection through the author__not_in parameter in WP_Query. Together they let an unauthenticated attacker run code on a target server. Searchlight Cyber withheld technical details in its initial wp2shell writeup on July 17 and published a public checker at wp2shell.com.

The situation has since escalated. WordPress shipped patched releases 7.0.2 and 6.9.5 with forced automatic updates on July 18, and security firm watchTowr reported the first signs of in-the-wild exploitation the same day, after public proof-of-concept code began circulating. The flaws affect WordPress 6.9.0-6.9.4 and 7.0.0-7.0.1, with the SQL injection component reaching back to 6.8.0.

The cost-and-speed figures come from Searchlight Cyber and have not been independently verified. What is not in dispute is the exposure: sites that have not taken the forced update are running code an automated exploit already knows how to reach.