OpenAI's AI Model Exposes Critical PHP Flaws Affecting Majority of the Web

OpenAI announced on October 21 that its GPT-5.6 Sol model, utilized under the Patch the Planet cybersecurity initiative, identified critical security vulnerabilities in the PHP programming language. Because PHP powers an estimated 70% of the internet, the discovery carries significant weight for global web infrastructure. The company reported that a coalition of researchers, maintainers, and government agencies worked over two months to resolve the issues, attributing the extended timeline to the complex engine-level changes required.

The core of the incident lies in the ability of AI-assisted researchers to convert a long-standing theoretical flaw into a practical threat. The PHP Foundation had been aware of the underlying issue since 2024 but struggled to identify a viable exploitation method or a fix. Human developers examined the code for two years without success. In contrast, researchers using OpenAI tools demonstrated working exploits within two weeks of beginning the audit, highlighting a sharp divergence between traditional human analysis and AI-accelerated security testing.

Two specific vulnerabilities, tracked as CVE-2026-109618 and CVE-2026-109619, were found in how PHP manages memory and handles errors. The first flaw occurs when the engine exhausts allocated memory, potentially leaving incomplete data that is later mistaken for valid. The second involves a fatal error causing the engine to point to freed memory, which subsequent requests can access or corrupt. Both flaws allow attackers to execute malicious code remotely without login credentials, potentially leading to full server compromise. They also risk exposing sensitive data such as passwords and API keys stored in server memory.

The severity of these bugs is underscored by their location in the PHP core engine, meaning websites are vulnerable even if their application code is secure. Both issues received a critical severity score of 9.2 out of 10. In response, the PHP Foundation released patched versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12. Older versions, including PHP 8.1 and earlier, will not receive official fixes, necessitating upgrades for long-term security. The coordinated release was moved up from October 22 to October 21 following premature disclosures on October 20 that sparked public concern.

OpenAI framed the incident as a demonstration of its 'Defender's Window' concept, a term introduced after its models previously hacked Hugging Face. This concept describes a narrow timeframe where organizations can leverage advanced AI to find and patch flaws before those same capabilities are widely adopted by malicious actors. The company argues that because powerful technology will inevitably be used by bad actors, there is a collective responsibility to proactively address exploitable weaknesses.

The discovery process involved a specific sequence of events beginning on July 30, 2026, when researchers from Calif, a California-based firm and founding partner of Patch the Planet, began auditing PHP at the Foundation's request. On August 11, Calif researcher Ryan A. identified the first exploitable vulnerability, with OpenAI researchers confirming it the same day. Later, the cybersecurity firm Wiz independently identified the second vulnerability. The combination of these efforts led to the current patches.

There is an open question regarding the practical impact of these flaws. Some argue that because the bugs required AI to be exploited, they would have remained harmless in a pre-AI threat landscape. OpenAI challenges this view by noting that Anthropic's Claude Mythos model recently identified thousands of zero-day vulnerabilities across major operating systems and browsers, suggesting that AI-driven threat discovery is becoming the norm rather than the exception.

For web developers and system administrators, the immediate priority is upgrading to the latest patched PHP versions to mitigate the risk. The incident serves as a case study in how AI is reshaping the cybersecurity landscape, compressing the timeline for vulnerability exploitation and forcing a faster response cycle for maintainers of critical infrastructure.

As AI models become more integrated into security workflows, the distinction between discovery and exploitation is blurring. This event may mark a shift in how the industry views legacy code, where previously dormant flaws can now be quickly weaponized, demanding continuous and automated vigilance from those who maintain the internet's foundational software.