AI Models Exploit Long-Overlooked PHP Flaws Affecting Most of the Web

A critical security gap that eluded human developers for two years was turned into a working exploit by AI-assisted researchers in just two weeks, exposing a vast portion of the internet to potential remote attacks. OpenAI announced on October 21 that its GPT-5.6 Sol model, deployed under the "Patch the Planet" cybersecurity initiative, identified severe vulnerabilities in PHP, the open-source language estimated to power 70% of all websites. The discovery highlights a growing tension in cybersecurity: while advanced AI tools can rapidly uncover and weaponize complex flaws, the window to fix them before malicious actors do is narrowing.

The timeline of the crisis reveals a long history of struggle. The PHP Foundation first became aware of the underlying flaw in 2024 but could not determine how to exploit it or fix it. In July 2026, the Foundation requested an audit using OpenAI tools. On August 11, researchers from the California-based firm Calif identified the exploitable vulnerability, with OpenAI researchers confirming it the same day. Shortly after, the security firm Wiz independently found a second, related flaw. These issues, tracked as CVE-2026-109618 and CVE-2026-109619, involve memory management and error handling errors that allow attackers to execute malicious code on servers without login credentials.

The technical severity of these bugs is high, with both receiving a critical score of 9.2 out of 10. The flaws exist within PHP’s core engine, meaning websites are vulnerable even if their specific application code is secure. Attackers can use these weaknesses to access sensitive data stored in server memory, such as passwords, authentication tokens, and API keys. In some cases, the vulnerabilities allow for complete control of affected web servers, making the threat particularly urgent for large-scale operations.

The release of patches was accelerated due to premature disclosures on October 20 that sparked public concern. Originally scheduled for October 22, the coordinated release of fixed versions PHP 8.2.35, 8.3.36, 8.4.27, and 8.5.12 was moved up to October 21. Older versions, including PHP 8.1 and below, will not receive official fixes. OpenAI states that all sites can now safely upgrade to remain secure, noting that the two-month coordination period involved major changes to the PHP engine and included input from government cybersecurity agencies.

OpenAI frames this incident as a prime example of the "Defender's Window," a concept the company introduced after its models hacked Hugging Face. This theory posits that there is a narrow timeframe where organizations can use frontier AI to strengthen defenses before those same capabilities become widely available to threat actors. While some argue that the flaws would have remained harmless without AI to exploit them, OpenAI maintains that proactive action is a collective responsibility. The company points to recent findings by Anthropic’s Claude Mythos, which identified thousands of zero-day vulnerabilities in major operating systems and browsers, as evidence that this new era of AI-driven security discovery is accelerating.

The situation underscores a shifting dynamic in software security. Human teams had examined the code for two years without resolution, whereas AI-wielding researchers demonstrated practical exploitation methods in a matter of weeks. As PHP maintainers and the broader tech industry digest this event, the focus remains on rapid patching and the strategic use of AI to close the gap between vulnerability discovery and public exploitation.