Two bugs sat inside PHP almost unnoticed for roughly twenty years. They weren't found by a team of security engineers but by an AI — actually by two, working independently. Wiz, a cybersecurity firm owned by Google, pointed its own AI agent at PHP's source code. Separately, a firm called Calif got a near-identical result using OpenAI's GPT-5.6 model, nicknamed Sol. Both landed on the same flaw, now called PHP2Shell and tracked as CVE-2026-109618 and CVE-2026-109619.
In plain terms, the bug lets an attacker remotely run arbitrary code on someone else's server without a password or login — a full takeover. PHP powers roughly seven out of ten websites whose language can be identified, including WordPress and thousands of government portals. Wiz itself called the find "among the most severe vulnerabilities ever discovered in PHP, and arguably in software history."
Wiz found the first bug on August 19 and reported it to the PHP team five days later. Developers quietly prepared a coordinated fix together with Cloudflare, AWS, Microsoft, Google Cloud, Automattic (which owns WordPress.com), Fastly, Akamai, Imperva and Vercel — a list that hints at how much of the internet depended on this patch. Public disclosure was planned for October 22, but details leaked online on October 20, forcing an early release: PHP shipped fixes on October 21 at 14:00 UTC for versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35, with backports for 8.1, 8.0 and 7.4.
While patches rolled out, several governments took parts of their services offline as a precaution. In the Netherlands, the Overheid.nl portal, the KNVB football federation's site, pages of the largest political parties and several hospitals went dark. Finland's Valtori pulled the state network offline, France put Interior Ministry sites into preventive maintenance, Denmark's Kombit school platform went down, and Japan's Sakura Internet disabled part of its database functions. The US NSA issued a rare public warning, and Wikipedia temporarily shut down part of its own systems.
What makes this story notable isn't just the bug itself, but who found it. Two companies, two different AI systems, the same discovery — months before any human reported it. Security researchers are now saying it plainly: if AI can dig up twenty-year-old holes in code the entire internet trusts, the same tools will eventually land in attackers' hands too, and then the clock will run in hours, not months.
