AI-Powered Research Uncovers Critical PHP Vulnerabilities

Researchers using advanced AI models have discovered two critical vulnerabilities in the PHP programming language, which underpins an estimated 70% of all websites. The flaws, collectively dubbed "PHP2Shell," allow an attacker to take complete control of a server without a password, prompting a global scramble among cybersecurity agencies and website administrators to apply emergency patches.

The two vulnerabilities, tracked as CVE-2026-109618 and CVE-2026-109619, both received a severity score of 9.2 out of 10. According to security researchers, a successful exploit would allow for unauthenticated remote code execution. This would enable attackers to steal sensitive data like passwords and API keys, install malware, or take a website offline entirely. The flaws affect a vast range of PHP versions, from 5.4 onward, covering more than a decade of releases used by millions of web properties, including government sites and major platforms like Facebook and Wikipedia.

The Role of AI in Uncovering a Hidden Threat

The discovery marks a significant moment for the use of artificial intelligence in cybersecurity. On August 11, 2026, the AI red-teaming firm Calif reported the first vulnerability after using OpenAI's GPT-5.6 Sol model to analyze PHP's source code. As part of an OpenAI initiative called "Patch the Planet," researchers tasked the AI with parsing the low-level memory allocation logic of the programming language. The model identified an unhandled exception state that could be triggered during out-of-memory operations, creating an exploitable security hole.

Thirteen days later, on August 24, the cybersecurity firm Wiz independently identified additional vulnerabilities using a multi-model toolchain that included Google's Gemini 4 Argon, OpenAI's GPT-5.6 Sol, and its own proprietary security framework. Both Wiz and OpenAI stated that frontier AI models were instrumental in turning long-dormant, complex bugs into workable exploits. In a statement, OpenAI noted that its AI-assisted researchers turned a two-year-old bug report that had stumped human developers into a working exploit in just two weeks. Wiz called the flaws "one of the most severe vulnerabilities discovered in PHP, and arguably in software history."

The underlying bugs are not new. The history of the flaws is disputed, with some reports suggesting they have existed in PHP's open-source codebase for as long as 14 years, or possibly even since 2004. The PHP Foundation acknowledged that one of the flaws was identified during an external security audit in 2024, but no action was taken at the time. The foundation attributed the delay to a "structural failure, caused by both the security team's decision process and its lack of resources."

A Coordinated Global Response

Following the private disclosures from Calif and Wiz in August, OpenAI coordinated a two-month effort involving PHP maintainers, security providers, and government agencies to develop and test patches. This culminated in a public announcement on Wednesday, October 21, 2026, when the PHP Foundation released emergency security updates. The foundation issued a severe warning, stating the vulnerabilities "sit in parts of PHP that nearly every request touches," and urged administrators to install updates immediately or take their sites offline.

Patches were released for PHP versions 8.2 through 8.5. The security team also provided minimal source code backports for older, unsupported versions 7.4, 8.0, and 8.1. The scale of the exposure is immense; the security firm Censys identified approximately 2.58 million internet-facing hosts running PHP at the time of disclosure, while an analysis by VirusTotal showed 5.1 million IP addresses.

Government cybersecurity agencies worldwide amplified the warning. The U.S. Cybersecurity and Infrastructure Security Agency (CISA), the National Security Agency (NSA), and the UK's National Cyber Security Centre all issued alerts urging organizations to prioritize patching internet-facing systems or to take them offline if they could not be updated. Following the disclosure, some government websites in Europe were temporarily taken down as a precautionary measure.

It remains unclear whether the vulnerabilities are being actively exploited by malicious actors. According to a report from Krebs on Security, some experts believe working exploits are already being used in the wild, and Cloudflare reported detecting some exploitation attempts. However, in its initial alert, CISA stated that it had not yet observed active exploitation.

What This Means for the Future of Cybersecurity

The PHP2Shell discovery is being compared to the 2021 Log4Shell vulnerability due to its widespread impact and the difficulty organizations face in patching all affected systems. Security experts warn that a "long tail of websites," particularly those managed by individuals or small organizations with limited resources, could remain vulnerable for weeks or months. Voldemaras Kadys, head of security at Mediatech, told reporters, "The potential impact of these vulnerabilities may be huge, as most websites run on PHP."

This event also serves as a powerful demonstration of AI's dual role in security. While AI helped defenders find and fix a critical flaw, OpenAI warned that the same capabilities are available to attackers. The company stated that it expects similar AI-driven security findings to "increase sharply in the coming months," creating a new kind of cat-and-mouse game between AI-powered defenders and AI-powered attackers. As Wiz noted, "Frontier AI models substantially reduced that barrier, turning a longstanding flaw into an exploit."

In response to the incident, OpenAI has proposed a five-stage process for handling critical vulnerabilities discovered by AI in widely used software. The framework, which grew out of the two-month response to PHP2Shell, is intended to serve as a model for coordinating disclosure and patching efforts for future high-impact discoveries. For now, the internet faces a race against time to secure millions of servers before attackers can widely operationalize the powerful exploits that AI helped uncover.