THE SCOREBOARD FOR WHAT'S NEXTAI + QUANTUM, SCORED. COMPARED. EXPLAINED.
AIQUANTUMSCORE.
← AI & Quantum News

AI / NEWS BRIEF

AI-assisted research exposes two critical flaws in PHP's core

PHP has issued emergency updates for two CVSS 9.2 vulnerabilities that can permit unauthenticated remote code execution, after independent AI-assisted discoveries by Calif and Wiz.

By AIQuantumScore · · Event 2026-10-21

Two critical core-engine flaws

The PHP project released emergency fixes for CVE-2026-109618 and CVE-2026-109619, two vulnerabilities rated 9.2 under CVSS 4.0. PHP says both can allow unauthenticated remote code execution in installations serving web requests, including PHP-FPM, Apache mod_php, php-cgi, FrankenPHP and LiteSpeed. Because the weaknesses sit in PHP's core rather than one application, secure WordPress, Drupal or Joomla application code does not remove the underlying exposure. Wiz also demonstrated process-memory disclosure that could expose passwords, tokens, API keys and certificates.

AI changed what was practical to exploit

OpenAI says Calif found a critical flaw using GPT-5.6 Sol through the Patch the Planet initiative, while Wiz independently discovered and reported both vulnerabilities. Wiz says its researchers used Wiz Atlas, Gemini 4 Argon and GPT-5.6 Sol, and that frontier models reduced the effort required to turn longstanding memory-safety defects into working demonstrations. PHP maintainers led the fixes, with researchers and infrastructure providers coordinating patches and protections before public disclosure.

Update, do not rely on partial mitigations

PHP identifies 8.5.12, 8.4.27, 8.3.36 and 8.2.35 as fixed supported releases and advises users of vendor-managed packages to obtain the corresponding update from their distributor. Versions 8.1 and older no longer receive normal PHP-project updates, although some long-term-support vendors provide backports and PHP published limited source backports for 7.4 through 8.1. The project's central warning is unambiguous: partial configuration changes can reduce some attack paths, but only an update is a complete solution.

Why it matters

OUR ANALYSIS

This incident is important beyond one software package. It provides concrete evidence that frontier models can make exploit chains practical when their complexity previously limited human researchers. That helps defenders uncover hidden risks, but it can also shorten the time between disclosure and widespread attack attempts. The discovery demonstrates a cybersecurity capability and coordination process; it is not a controlled model benchmark and should not be converted into an AIQuantumScore or REVAL result.

More to follow