70% of the Internet: The PHP Flaw That AI Exposed in Two Weeks

An estimated seventy percent of the internet is built on PHP, and for the first time, a critical flaw in that foundational code has been fully weaponized by artificial intelligence. OpenAI announced on October 21 that its GPT-5.6 Sol model, operating under the company’s “Patch the Planet” cybersecurity initiative, identified severe vulnerabilities in the PHP engine. These flaws, which allow for remote code execution without login credentials, were not just discovered but converted into working exploits in a span of two weeks, a process that had eluded human experts for two years.

The scope of the threat is defined by two specific Common Vulnerabilities and Exposures: CVE-2026-109618 and CVE-2026-109619. Both issues stem from how PHP manages computer memory and handles errors. The first vulnerability occurs when the engine runs out of allocated memory during a request, potentially leaving behind incomplete internal data that is later mistakenly treated as valid. The second involves a fatal error that leaves PHP pointing to memory that has already been freed, allowing subsequent requests to access or corrupt that space. Both flaws were assigned a critical severity score of 9.2 out of 10, reflecting the high risk they pose to server infrastructure.

The remediation process was complex, requiring major changes to the core PHP engine. Consequently, the PHP Foundation released patched versions—8.2.35, 8.3.36, 8.4.27, and 8.5.12—to address the issues. Older versions, specifically PHP 8.1 and below, will not receive official fixes. While the coordinated release was originally scheduled for October 22, it was brought forward to October 21 after premature disclosures on October 20 sparked public outcry. OpenAI confirmed that any site using PHP can safely upgrade to these new versions to secure their infrastructure.

The discovery was a collaborative effort driven by the Patch the Planet program, of which California-based cybersecurity firm Calif is the founding partner. At the request of the PHP Foundation, Calif researchers began auditing the code using OpenAI tools on July 30, 2026. The underlying flaw had been known to the Foundation since 2024, but no one could identify a method to exploit it or a way to fix it. On August 11, Calif researcher Ryan A. identified the exploitable vulnerability, with OpenAI researchers confirming the finding the same day. Separately, another firm, Wiz, independently identified an additional vulnerability in the same area.

The implications of these bugs extend beyond simple code execution. Researchers demonstrated that the flaws could expose sensitive data stored in server memory, including passwords, authentication tokens, and API keys. Because the vulnerabilities reside in PHP’s core engine, websites are vulnerable even if their specific application code is perfectly secure. This distinguishes the incident from typical web application flaws, which usually require a mistake in how a developer writes code. Here, the risk is inherent to the engine itself.

OpenAI frames this incident as a prime example of what it calls the “Defender's Window.” This concept, introduced after the company’s models hacked Hugging Face, describes a narrow timeframe where organizations can use frontier AI to find and patch defenses before malicious actors gain access to similar capabilities. The company argues that because powerful technology will inevitably be used by bad actors, it is a collective responsibility to act proactively. This stance is supported by recent reports that Anthropic’s Claude Mythos identified thousands of zero-day vulnerabilities in major operating systems and browsers, suggesting the trend of AI-driven discovery is accelerating.

The timeline highlights the shift in cybersecurity dynamics. Human developers and security teams had examined the underlying code since 2024 but could not fully resolve the issues, allowing them to persist for two years. In contrast, AI-wielding researchers demonstrated practical exploitation methods within two weeks of the audit beginning. While some may argue that the flaws would not have been exploitable without AI, rendering them effectively harmless, OpenAI challenges this view by emphasizing the inevitability of AI adoption in offensive security.

For California readers and web developers across the state, the urgency is clear. The two-month coordination period between researchers, maintainers, and government agencies was necessary due to the complexity of the engine changes. However, the window for safe upgrading is now open. As AI models continue to probe the foundations of the internet, the ability to identify and patch core language vulnerabilities before they become widespread attack vectors is becoming a critical component of digital defense.