AI Finds Critical Flaws in PHP Core

OpenAI announced on October 21 that researchers under its “Patch the Planet” cybersecurity initiative used the GPT-5.6 Sol model to identify critical vulnerabilities in PHP, the open-source language powering an estimated 70% of the internet. The discovery triggered a coordinated effort involving PHP maintainers, security providers, and government agencies to develop fixes over a two-month period, a timeline OpenAI attributed to the significant engineering changes required within the PHP engine.

The urgency of the response was driven by the scope of the affected infrastructure. The PHP Foundation released patched versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12 to address the issues, while noting that versions 8.1 and older would not receive official updates. Although the coordinated release was initially planned for October 22, it was accelerated to October 21 after premature disclosures on October 20 sparked public concern. OpenAI stated that all PHP-based sites can now safely upgrade to secure their systems.

The technical core of the issue involves two distinct flaws in how PHP manages memory and handles errors, tracked as CVE-2026-109618 and CVE-2026-109619. The first vulnerability occurs when the engine runs out of allocated memory during a request, potentially leaving incomplete internal data that is later misinterpreted as valid. The second arises when a fatal error leaves the system pointing to already-freed memory, allowing subsequent requests to access or corrupt that space. Both flaws carry a critical severity score of 9.2 out of 10 and reside in the core engine, meaning they can affect websites even if the application code itself is secure.

The timeline of the discovery highlights the shifting dynamics of modern security research. The PHP Foundation had been aware of the underlying flaw since 2024 but had failed to find an exploitation method or a fix for two years. Audits began on July 30, 2026, at the Foundation's request, involving Calif, a California-based cybersecurity firm and founding partner of the Patch the Planet program. On August 11, Calif researcher Ryan A. identified the exploitable vulnerability, a finding matched by OpenAI researchers on the same day. The firm Wiz independently identified the second vulnerability during this period.

OpenAI frames this incident as a prime example of its “Defender's Window” concept, a term introduced after its models previously hacked Hugging Face. The concept describes a narrow timeframe where organizations can use frontier AI to strengthen defenses before malicious actors gain access to similar capabilities. By demonstrating that AI-wielding researchers could create working exploits in just two weeks—whereas human teams had been stuck for two years—OpenAI argues that proactive action is a collective responsibility. The company contends that while some may view these flaws as harmless without AI, the inevitability of AI-driven attacks necessitates immediate remediation.

The implications extend beyond simple patching, as the researchers demonstrated that these flaws could expose sensitive data stored in server memory, such as passwords, authentication tokens, and API keys. Attackers could potentially exploit these weaknesses remotely without login credentials to execute malicious code and take control of affected servers. OpenAI considers this one of the biggest cybersecurity discoveries of 2026 due to PHP's ubiquity, reinforcing the argument that advanced AI models are now essential tools for uncovering hidden weaknesses in foundational software.