AI Finds Two-Year-Old PHP Flaw Affecting 70% of Web
A critical bug that stumped human experts for two years was turned into a working exploit by AI researchers in just two weeks, forcing an emergency patch for a vast portion of the internet.
By Amelia Gupta ·
Researchers have successfully weaponized a two-year-old bug in PHP, the open-source language that powers an estimated 70% of the internet, after an AI model identified a viable attack path that human developers had missed. The discovery, announced on Tuesday, highlights a significant shift in cybersecurity where artificial intelligence has begun to outpace traditional manual code audits in finding and exploiting deep-seated vulnerabilities. OpenAI stated that its GPT-5.6 Sol model was central to uncovering these critical flaws, which allow attackers to execute malicious code on affected servers without needing login credentials.
The vulnerabilities, tracked as CVE-2026-109618 and CVE-2026-109619, reside in PHP's core engine rather than in application-specific code. This location is particularly concerning because it means websites can be compromised even if their own code is secure. The issues involve memory management and error handling. Specifically, when PHP runs out of memory during a request, it may leave incomplete data behind that the engine later treats as valid. A second flaw causes fatal errors to leave PHP pointing to freed memory, which subsequent requests can then access or corrupt. Both issues received a critical severity score of 9.2 out of 10.
The timeline of the discovery reveals a stark contrast in efficiency between human and AI-assisted analysis. The PHP Foundation had been aware of the underlying flaw since 2024 but failed to identify a method for exploitation or a fix. In July 2026, researchers from Calif, a founding partner of OpenAI's Patch the Planet initiative, began auditing the code using AI tools. On August 11, both Calif and OpenAI researchers independently identified the exploitable vulnerability. Within two weeks, they had developed working exploits, a feat that human teams had not achieved in the previous two years.
Following the discovery, a coordinated response involving OpenAI, PHP maintainers, security providers, and government agencies worked for two months to develop patches. The process required major changes to the PHP engine, explaining the lengthy timeline. The PHP Foundation released fixed versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12. Notably, PHP 8.1 and older versions will not receive official fixes, leaving legacy systems potentially exposed. The coordinated release was originally scheduled for Wednesday but was moved up to Tuesday after premature disclosures on Monday sparked public concern.
OpenAI framed this incident as a prime example of its Defender's Window concept. Introduced after its models hacked Hugging Face, the theory suggests there is a narrow timeframe where organizations can use frontier AI to find and patch defenses before those same capabilities become widely available to malicious actors. The company argues that relying on the idea that flaws are harmless until exploited is a dangerous strategy, especially given that other AI models have recently identified thousands of zero-day vulnerabilities in major operating systems and browsers.
The researchers also demonstrated that the flaws could expose sensitive data stored in server memory, such as passwords and API keys. While some may argue that the vulnerabilities would never have been exploitable without AI, making them effectively inert, OpenAI maintains that proactive defense is a collective responsibility. With the patches now available, every site using PHP is urged to upgrade immediately to secure its infrastructure against these newly revealed threats.
