OpenAI Finds Critical PHP Flaws That Threaten Most Websites

Seventy percent of the internet runs on PHP, and that vast digital infrastructure was nearly left exposed to remote attacks. On Oct. 21, OpenAI announced that its GPT-5.6 Sol model helped uncover critical security vulnerabilities in the open-source language. The discovery highlights a significant shift in cybersecurity, where artificial intelligence is identifying weaknesses that human developers had overlooked for years. The company described the incident as one of the most significant cybersecurity findings of 2026 due to the sheer scale of PHP usage.

The vulnerabilities, tracked as CVE-2026-109618 and CVE-2026-109619, stem from how PHP manages memory and handles errors. The first flaw occurs when the system runs out of allocated memory during a request. This can leave behind incomplete internal data that the engine later mistakes for valid information. The second issue arises when a fatal error causes the system to point to memory that has already been freed. In both cases, attackers could potentially access or corrupt that memory without needing login credentials, allowing them to execute malicious code and take control of servers.

Researchers also demonstrated that these flaws could expose sensitive data stored in server memory. This includes passwords, authentication tokens, and API keys. The problems exist in the core engine of PHP, meaning websites can be vulnerable even if their specific application code is secure. Both vulnerabilities received a critical severity score of 9.2 out of 10, underscoring the high risk they posed to online services.

The discovery process began in late July 2026 when researchers from Calif, a California-based cybersecurity firm, started auditing PHP at the request of the PHP Foundation. The Foundation had known about the underlying flaw since 2024 but could not find a way to exploit it or fix it. On Aug. 11, a Calif researcher identified the exploitable vulnerability, with OpenAI researchers doing the same that day. Shortly after, another firm, Wiz, independently found an additional related vulnerability.

OpenAI stated that human teams had examined the code for two years without resolving the issue. However, the AI-assisted researchers turned the bug into a working exploit within two weeks. The company argues that while the flaws might not have been exploitable without AI, the rapid advancement of technology means bad actors will eventually have access to similar tools. This supports the concept of a "Defender's Window," a narrow timeframe where organizations must use advanced AI to patch defenses before attackers do.

Following the discovery, a coalition of researchers, maintainers, and government agencies worked for two months to fix the issues. The process was lengthy because the solution required major changes to the PHP engine. The PHP Foundation released patched versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12 to address the problems. Sites running PHP 8.1 or older versions will not receive official fixes.

The coordinated release of these patches was originally scheduled for Oct. 22. However, it was moved up to Oct. 21 after premature disclosures on Oct. 20 caused public concern. OpenAI says that any site using PHP can now safely upgrade its version to stay secure. The company emphasized that proactive action is a collective responsibility to address potential flaws before they become widespread threats.

This event follows a similar finding by Anthropic, whose Claude Mythos model recently identified thousands of zero-day vulnerabilities in major operating systems and web browsers. OpenAI sees the PHP discovery as a proof of concept for its Defender's Window idea. The company introduced this concept after its models hacked Hugging Face, illustrating how frontier AI models can be used to strengthen defenses before those same capabilities become available to malicious actors.