OpenAI President Greg Brockman announced that the company led a two-month coalition to mitigate PHP2Shell, a severe vulnerability in the PHP programming language discovered using OpenAI’s models under its Patch the Planet initiative. The flaws allowed unauthorized code execution on servers running PHP, which powers approximately 70% of websites with known server-side languages. Security firm Calif identified a version of the bug in August using GPT-5.6 Sol, while OpenAI found a fuller version in September; cloud security firm Wiz independently discovered both flaws. The disclosure process faced complications when details leaked early, prompting PHP maintainers to release fixes on October 21 instead of the planned October 22 date.
The discovery triggered immediate defensive actions, causing Finland, the Netherlands, and France to take government websites offline on Wednesday to prevent exploitation. Brockman emphasized the need for authorized defenders, such as governments, to scan for vulnerable servers and potentially patch them remotely in extreme cases. He argued that defenders require tighter operational cycles than attackers, advocating for national priorities to ensure access to capable cyber models. OpenAI has committed $1 billion in subsidized access to Daybreak, its AI program for critical infrastructure defense, and is addressing other significant bugs, including a Node.js remote code execution flaw.
This incident marks a pivotal shift in cybersecurity dynamics, where AI tools are not just assisting human analysts but actively identifying zero-day vulnerabilities at scale. The rapid response by European governments highlights the tangible risk posed by such discoveries, forcing immediate operational shutdowns to mitigate potential widespread exploitation. It underscores the urgency for defenders to adopt advanced AI capabilities to match the speed and sophistication of threats, moving beyond traditional manual auditing processes.
However, the reliance on AI for vulnerability discovery introduces complex governance challenges. The premature leak of details despite a coordinated disclosure effort reveals the fragility of current information-sharing protocols among private entities and public sectors. As OpenAI positions itself as a key defender through initiatives like Patch the Planet, questions arise regarding the balance between proactive security scanning and privacy concerns, especially given past incidents involving unauthorized data access. The call for governments to remotely patch servers suggests a future where automated intervention becomes standard, necessitating clear legal frameworks to define authorized defensive actions.
