Seventy Percent of the Web: AI Agents Turn Stalled PHP Bugs into Critical Threats

An estimated seventy percent of the internet runs on PHP, a ubiquity that has just turned the open-source language into the central stage for one of the most significant cybersecurity events of 2026. On October 21, OpenAI confirmed that researchers operating under its “Patch the Planet” initiative discovered critical vulnerabilities in the language’s core engine. The finding was made possible by GPT-5.6 Sol, an AI model that helped transform a known but unresolved bug into a set of working exploits that threaten servers worldwide.

The vulnerabilities, identified as CVE-2026-109618 and CVE-2026-109619, stem from fundamental issues in how PHP manages computer memory and processes errors. One flaw occurs when the engine exhausts its allocated memory during a request, potentially leaving behind incomplete data that is later misinterpreted as valid. The second involves a fatal error that leaves the system pointing to memory that has already been freed. These weaknesses allow attackers to remotely execute malicious code without login credentials, potentially seizing control of affected websites.

What makes this incident particularly striking is the timeline of the discovery. The PHP Foundation had been aware of the underlying flaw since 2024, yet human developers were unable to identify a method for exploitation or a fix. It took two years for the bug to remain dormant in a way that stumped traditional security teams. Once researchers from the California-based firm Calif began auditing the code with OpenAI tools on July 30, the situation accelerated. By August 11, both Calif and OpenAI researchers identified the exploitable nature of the vulnerability, a process that took only two weeks to demonstrate practical attack methods.

The PHP Foundation released patched versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12 to address the issues, with older versions left without official fixes. The coordinated release was originally scheduled for October 22 but was moved up to October 21 following premature disclosures on October 20 that sparked public concern. The company attributed the two-month period required to fix the issues to the complexity of the necessary changes, which required major modifications to the PHP engine. Both vulnerabilities received a critical severity score of 9.2 out of 10.

OpenAI framed this event as a demonstration of its “Defender’s Window” concept, a narrow timeframe where organizations can use advanced AI to strengthen defenses before those same capabilities become widely available to threat actors. The company noted that AI tools also revealed the flaws could expose sensitive data in server memory, including passwords and API keys. This development follows recent reports that Anthropic’s Claude Mythos identified thousands of zero-day vulnerabilities in major operating systems and browsers. For now, the incident underscores the shifting balance of power in cybersecurity, where AI-assisted researchers are moving from finding theoretical bugs to crafting real-world exploits with unprecedented speed.

The discovery highlights a broader trend in the industry: the gap between human and machine capabilities in identifying complex software weaknesses. While human experts had examined the code for two years without resolution, AI-wielding researchers found practical exploitation paths in a matter of weeks. As OpenAI argues, the responsibility to act proactively against these AI-driven threats is a collective one. The rapid patching of PHP serves as a case study in how the Defender’s Window can be utilized to secure critical infrastructure before the next wave of AI-assisted attacks emerges.