Patch the Planet: coordinating fixes for AI-found vulnerabilities

Summary

OpenAI disclosed how it used AI (GPT-5.6 Sol) to find critical PHP vulnerabilities and then worked for two months with open-source maintainers, security firms, and government agencies to prepare fixes and disclosure. Drawing on the case, it proposed a five-stage Patch the Planet process covering validation, advance information sharing, government coordination, distribution, and publication, and committed $1 billion in subsidized Daybreak access for defenders with limited security resources. Wiz independently found the flaws, PHP maintainers led the fixes, and OpenAI itself calls this one of the most impactful security issues in recent memory.

Why it matters for our work

A 'defender's window' where frontier AI finds flaws before attackers has moved from concept to documented case. If such findings multiply, patching speed becomes a core competency for development and operations teams.

Translated from the Korean original. Summaries may be translated and edited. Commentary reflects our perspective; forecasts remain the source’s views.