OpenAI-led team patches critical PHP flaws affecting seven in ten websites

The PHP Foundation pulled forward the release of critical security patches to October 21, one day ahead of schedule, after premature disclosures triggered a wave of public concern on October 20. The urgent move addresses two severe vulnerabilities in the PHP engine, an open-source programming language that underpins an estimated 70% of the internet. By accelerating the coordinated release, the foundation aimed to mitigate the risk of widespread exploitation before the details became fully public. The updated versions, PHP 8.2.35, 8.3.36, 8.4.27, and 8.5.12, are now available for download, while older iterations including PHP 8.1 and prior versions remain without official fixes.

The discovery was driven by an AI-assisted audit conducted under OpenAI’s “Patch the Planet” cybersecurity initiative. Researchers utilized the GPT-5.6 Sol model to analyze the PHP codebase, a task that had stumped human developers for two years. The foundation had been aware of the underlying flaw since 2024 but lacked a method to exploit it or a viable fix. The involvement of AI changed the dynamic, allowing the team to convert a theoretical weakness into a practical threat model within a significantly shorter timeframe than traditional manual auditing would have allowed.

The technical core of the issue involves two distinct but related memory management errors. The first, tracked as CVE-2026-109618, occurs when PHP exhausts its allocated memory during a request. This can leave incomplete internal data in the system, which the engine may later incorrectly interpret as valid. The second flaw, CVE-2026-109619, arises when a fatal error leaves the system pointing to memory that has already been freed. This dangling pointer allows subsequent requests to access or corrupt that memory space. Both vulnerabilities carry a critical severity score of 9.2 out of 10, reflecting the high risk they pose to server integrity.

The financial and operational stakes are significant because these flaws reside in PHP’s core engine. This means that websites can be vulnerable even if their specific application code is secure. Attackers can exploit these weaknesses remotely without requiring login credentials, potentially executing malicious code on affected servers. Furthermore, the researchers demonstrated that the bugs could expose sensitive data stored in server memory, including passwords, authentication tokens, and API keys. For organizations relying on PHP, this represents a direct threat to data confidentiality and system control.

The timeline of the discovery highlights the speed of AI-driven security research. The audit began on July 30, 2026, at the request of the PHP Foundation. On August 11, researchers from Calif, a California-based cybersecurity firm and founding partner of the initiative, first identified the exploitable vulnerability. OpenAI researchers confirmed the finding the same day. Shortly after, another cybersecurity firm, Wiz, independently identified the additional vulnerability. The coordinated effort to fix the issues took two months, a duration OpenAI attributes to the complexity of the necessary changes to the PHP engine.

OpenAI frames this incident as a prime example of its “Defender’s Window” concept. Introduced after its models interacted with Hugging Face, the concept describes a narrow timeframe where organizations can use advanced AI models to find and patch vulnerabilities before those same capabilities become widely available to malicious actors. The company argues that since powerful technology will inevitably be used by bad actors, there is a collective responsibility to act proactively. This approach contrasts with the view that flaws not yet exploitable by humans are effectively harmless, a stance OpenAI challenges by pointing to the speed with which AI can generate exploits.

The broader market impact is underscored by the ubiquity of PHP. With nearly three-quarters of the web running on this language, the vulnerability affects a vast ecosystem of businesses, from large enterprises to small e-commerce sites. The fact that the bug persisted for two years without a fix underscores the limitations of traditional security auditing methods. The successful use of GPT-5.6 Sol to uncover and help resolve the issue signals a shift in how critical infrastructure security is managed. It suggests that AI tools are becoming essential for identifying deep-seated flaws in foundational software components that have resisted human detection for extended periods.

The incident also serves as a data point in the ongoing debate about the role of AI in cybersecurity. While some argue that the vulnerabilities were only a problem because AI could exploit them, OpenAI maintains that the proactive use of these tools is necessary to stay ahead of the curve. The company notes that other AI models, such as Anthropic’s Claude Mythos, have recently identified thousands of zero-day vulnerabilities in major operating systems and browsers. This trend suggests that the pace of discovery is accelerating, and the window for defenders to patch issues is narrowing. For businesses, the message is clear: the cost of inaction is rising as AI-driven attacks become more sophisticated and widespread.

The coordinated response involved a coalition of researchers, PHP maintainers, security providers, and government cybersecurity agencies. This multi-stakeholder approach was crucial in ensuring that the fixes were robust and that the release was managed effectively to minimize disruption. The incident highlights the increasing importance of collaboration between tech firms, open-source foundations, and security specialists in managing the risks posed by emerging technologies. As AI continues to evolve, the ability to identify and patch vulnerabilities quickly will become a key competitive advantage for organizations seeking to secure their digital infrastructure.