AI Solvers Crack PHP Security Code That Stumped Humans for Two Years

The web runs on a specific infrastructure, and much of it relies on a single programming language that has long been a target for security concerns. On Wednesday, OpenAI revealed that its advanced AI models have identified a pair of critical vulnerabilities in PHP, the open-source language that underpins an estimated 70% of the internet.

The discovery was made by researchers operating under OpenAI’s “Patch the Planet” cybersecurity initiative, who utilized the GPT-5.6 Sol model to audit the codebase. This effort, conducted in collaboration with Calif, a California-based cybersecurity firm and founding partner of the program, began at the end of July at the specific request of the PHP Foundation. The foundation had been aware of an underlying issue since 2024 but lacked the tools to determine how it could be exploited or how to fix it.

The first key takeaway is that human developers had been stuck on this problem for two years. While the PHP Foundation knew a flaw existed, they could not identify a practical exploitation method or a viable solution. It was not until August 11 that Calif researcher Ryan A. identified an exploitable path, a finding that was immediately corroborated by OpenAI’s own research team on the same day. This highlights a significant shift in how deep codebase flaws are detected, moving from a months-long human struggle to a rapid AI-assisted breakthrough.

The second major development involves the specific nature of the two identified bugs, tracked as CVE-2026-109618 and CVE-2026-109619. Both issues stem from how PHP handles computer memory and error processing. The first occurs when the system runs out of allocated memory during a request, potentially leaving behind incomplete data that the engine later mistakes for valid information. The second involves a fatal error causing the system to point to memory that has already been freed, allowing subsequent requests to access or corrupt that space. These flaws reside in the core engine, meaning they affect websites even if the application code itself is secure.

The third critical point is the speed with which the AI tools turned a theoretical bug into a working exploit. According to OpenAI, researchers using AI demonstrated practical ways to exploit the vulnerabilities within just two weeks of identification. The tools also showed that attackers could potentially expose sensitive data stored in server memory, such as passwords and API keys, without needing login credentials. Both vulnerabilities were assigned a critical severity score of 9.2 out of 10, underscoring the high risk they posed to global infrastructure.

Contextually, the release of the fixes was more urgent than planned. The coordinated release of patched versions, PHP 8.2.35, 8.3.36, 8.4.27, and 8.5.12, was originally set for Thursday but was moved up to Wednesday due to premature disclosures on Tuesday that sparked public concern. The PHP Foundation clarified that versions 8.1 and older will not receive official patches, meaning older systems remain at higher risk. OpenAI coordinated with maintainers, security providers, and government agencies over the past two months to finalize these complex changes to the PHP engine.

OpenAI is framing this incident as a prime example of its “Defender’s Window” concept. This theory suggests there is a narrow timeframe where organizations can use frontier AI models to strengthen defenses before those same capabilities become widely available to threat actors. The company argues that because powerful technology will inevitably be used by bad actors, it is a collective responsibility to proactively address exploitable flaws. This follows a similar claim that Anthropic’s Claude Mythos recently identified thousands of zero-day vulnerabilities in major operating systems and browsers.

Looking ahead, the immediate action for web administrators is clear: upgrade to the latest patched versions of PHP to secure their servers. While some may argue that the flaws would have remained harmless without the AI tools to exploit them, OpenAI maintains that proactive defense is necessary. As AI-driven security audits become more common, the pace of patching and vulnerability discovery is likely to accelerate, requiring developers and system administrators to stay agile in their response to emerging threats.