OpenAI Reports AI Model Exposed Critical PHP Flaws Affecting Majority of Web

OpenAI announced on October 21 that researchers operating under its Patch the Planet initiative utilized its GPT-5.6 Sol model to identify critical security vulnerabilities in PHP, the open-source language powering approximately 70% of the internet. Following the discovery, the company coordinated with PHP maintainers, security providers, and government agencies to address the issues over a two-month period. The extended timeline was attributed to the complexity of the task, which necessitated significant modifications to the PHP engine.

The PHP Foundation issued patched versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12 to remediate the vulnerabilities. Older versions, including PHP 8.1, will not receive official fixes. Although the coordinated release was initially scheduled for October 22, it was accelerated to October 21 due to premature disclosures on October 20 that triggered public concern. OpenAI stated that all sites utilizing PHP can now safely upgrade their versions to maintain security.

The incident, which OpenAI considers one of the most significant cybersecurity discoveries of 2026, illustrates the company’s concept of the Defender’s Window. This term describes a narrow timeframe where organizations can leverage frontier AI models to strengthen defenses before similar capabilities become widely available to threat actors. The concept was introduced after OpenAI models previously interacted with Hugging Face. The company argues that proactive remediation is a collective responsibility, citing recent findings by Anthropic’s Claude Mythos regarding zero-day vulnerabilities in major operating systems and web browsers.

Auditing of PHP began on July 30, 2026, at the PHP Foundation’s request, involving researchers from Calif, a California-based cybersecurity firm and founding partner of the Patch the Planet program. The Foundation had been aware of the underlying flaw since 2024 but lacked an exploitation method or fix. On August 11, Calif researcher Ryan A. identified an exploitable vulnerability, followed by OpenAI researchers the same day. Separately, the cybersecurity firm Wiz independently identified an additional vulnerability.

The two vulnerabilities, tracked as CVE-2026-109618 and CVE-2026-109619, stem from flaws in PHP’s memory management and error handling. The first occurs when PHP exhausts allocated memory during a request, potentially leaving incomplete internal data that the engine later treats as valid. The second involves a fatal error causing PHP to point to freed memory, allowing subsequent requests to access or corrupt it. Both received a critical severity score of 9.2 out of 10. Because the flaws reside in the core engine, websites may be affected even if their application code is secure. Under specific conditions, attackers can exploit these weaknesses remotely without login credentials to execute malicious code.

AI-assisted researchers demonstrated that the flaws could expose sensitive server memory data, including passwords, authentication tokens, and API keys. They showed practical exploitation methods within two weeks, a timeframe significantly shorter than the two years human developers spent examining the code without fully resolving the issues. While some might argue the flaws were harmless without AI assistance, OpenAI maintains that advanced technology will inevitably be used by bad actors, necessitating immediate defensive action.

In the coming weeks, web administrators are expected to prioritize upgrading their PHP installations to the latest patched versions. OpenAI continues to promote the Defender’s Window framework as a strategic approach for organizations facing rapid advancements in AI-driven cybersecurity capabilities.