AI Models Exploit PHP Flaws That Stumped Humans for Two Years

Patched versions of PHP are officially available to fix two critical security holes that could affect roughly 70% of all websites. The updates, released on October 21, 2026, address vulnerabilities in the language’s core engine that allow remote attackers to execute malicious code without needing login credentials. This rapid release was pushed forward from its scheduled date after premature disclosures sparked public concern just days earlier.

To understand why this matters, it helps to look back at how these flaws were found. The PHP Foundation had known about the underlying issue since 2024, but human developers could not figure out how to exploit it or how to fix it. That changed in July when the cybersecurity firm Calif began auditing PHP using OpenAI’s tools under a program called Patch the Planet. By August 11, researchers had identified the first exploitable weakness, and OpenAI’s team confirmed it the same day. A separate firm, Wiz, later found a second related vulnerability.

Both flaws, tracked as CVE-2026-109618 and CVE-2026-109619, involve how PHP handles memory and errors. The first occurs when the system runs out of memory during a request, leaving behind incomplete data that the engine mistakenly treats as valid. The second happens when a fatal error causes PHP to point to memory that has already been freed, allowing later requests to access or corrupt it. Both issues received a critical severity score of 9.2 out of 10 because they exist in the core engine, meaning even secure application code can be compromised.

Using AI assistance, researchers demonstrated that these bugs could be exploited within two weeks to run malicious code and expose sensitive data like passwords and API keys. This stood in sharp contrast to the two years human teams struggled with the same code. OpenAI highlighted this incident as a prime example of what it calls the Defender's Window. This concept suggests there is a narrow timeframe where organizations can use advanced AI to find and patch security gaps before bad actors gain access to the same capabilities.

The company argues that since powerful technology will inevitably be used by attackers, there is a collective responsibility to fix exploitable flaws proactively. This view follows recent reports that other AI models have also identified thousands of zero-day vulnerabilities across major operating systems and browsers. For website owners, the immediate step is to upgrade to one of the patched versions. The PHP Foundation has released fixes for versions 8.2, 8.3, 8.4, and 8.5. Sites running PHP 8.1 or older will not receive official patches for these specific issues.

The coordinated fix involved a coalition of researchers, maintainers, and government agencies working over two months to implement the changes. The complexity required major modifications to the PHP engine, which is why the process took so long. Now that the patches are out, OpenAI states that any site using PHP can safely upgrade to secure itself against these remote attacks. This event marks a significant moment in 2026 cybersecurity, illustrating how AI is accelerating both the discovery and the remediation of critical software flaws.