AI finds two-year-old PHP bug that could hit three-quarters of the web

Approximately 70 percent of the internet runs on PHP, an open-source programming language that has been found to contain two critical security vulnerabilities. These flaws, which allow for remote code execution, have now been patched after a discovery process that highlighted the growing role of artificial intelligence in cybersecurity. The incident has been described as one of the most significant cybersecurity findings of 2026 due to the sheer scale of the software’s usage.

The vulnerabilities, tracked as CVE-2026-109618 and CVE-2026-109619, stem from errors in how PHP manages memory and handles system errors. The first issue arises when the system runs out of allocated memory, potentially leaving incomplete data that the engine later interprets as valid. The second involves a fatal error that leaves the system pointing to memory that has already been freed, allowing subsequent requests to corrupt it. Both issues received a critical severity score of 9.2 out of 10, meaning attackers could potentially take control of servers without needing login credentials.

The underlying flaw had been known to the PHP Foundation since 2024, but human developers were unable to determine how it could be exploited or how to fix it. The situation changed on July 30, 2026, when the PHP Foundation requested an audit using OpenAI’s tools. Researchers from Calif, a California-based cybersecurity firm and founding partner of OpenAI’s “Patch the Planet” initiative, began the audit using the GPT-5.6 Sol model. By August 11, Calif researcher Ryan A. had identified an exploitable vulnerability, and OpenAI researchers confirmed findings the same day. The firm Wiz independently identified the second vulnerability during the same period.

In a statement, OpenAI noted that while human teams struggled to resolve the code issues for two years, AI-assisted researchers developed working exploits within two weeks. The company framed this event as an example of its “Defender’s Window” concept, a narrow timeframe where organizations can use advanced AI to patch defenses before malicious actors gain similar capabilities. OpenAI argues that since powerful technology will inevitably be used by bad actors, there is a collective responsibility to proactively address exploitable flaws.

Following the discovery, a coalition of researchers, PHP maintainers, security providers, and government agencies worked for two months to develop a solution. The timeline was extended by the complexity of the task, which required major changes to the PHP engine. Although the coordinated release was originally planned for October 22, it was moved up to October 21 after premature disclosures on October 20 caused public concern. The PHP Foundation has released patched versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12 to address the issues.

The vulnerabilities also posed a risk to sensitive data, as researchers demonstrated that they could expose information stored in server memory, including passwords, authentication tokens, and API keys. Because the flaws exist in PHP’s core engine, websites are vulnerable even if their specific application code is secure. This distinction makes the issue particularly pressing for site administrators who may assume their security posture is robust based on their own code audits.

OpenAI stated that all sites using PHP can now safely upgrade to the new versions to remain secure. The company highlighted the incident as proof that AI models can uncover critical issues that have evaded human detection for extended periods. This follows recent reports from Anthropic, whose Claude Mythos model identified thousands of zero-day vulnerabilities in major operating systems and web browsers, further cementing the trend of AI-driven security discoveries.

Site administrators and developers are advised to check their PHP versions immediately. Versions 8.1 and older will not receive official fixes for these specific vulnerabilities, meaning users of those versions must plan for upgrades or apply workarounds. The next step for the broader web ecosystem is a mass migration to the patched versions to close the window of exposure before potential attacks are widely launched.

The incident underscores a shifting dynamic in software security, where the speed of AI-assisted discovery outpaces traditional human-led audits. As more organizations integrate AI into their security protocols, the pressure to patch critical infrastructure will likely increase. For now, the coordinated release marks the end of the immediate crisis, but the discovery serves as a reminder that long-standing code can hide significant risks.