OpenAI says GPT-5.6-Sol helped find the PHP flaws

OpenAI says its GPT-5.6 Sol model helped researchers find two critical security flaws in PHP, the programming language that runs an estimated 70% of websites. The vulnerabilities, known together as PHP2Shell, let an attacker run code on a web server without logging in. The discovery forced PHP's maintainers to ship emergency patches this week.

Security firm Calif found the first flaw in August with help from GPT-5.6 Sol during an audit of PHP's source code that OpenAI funded through its Patch the Planet initiative. Calif and OpenAI both flagged the same underlying problem on August 11. Security firm Wiz, which is part of Google, separately found both flaws using its own Wiz Atlas platform together with Google's Gemini 4 Argon and GPT-5.6 Sol. PHP's advisory credits researchers from Calif, Wiz and OpenAI. The PHP Foundation says the root cause, a memory-corruption bug tied to how PHP handles requests that hit its memory limit, was first flagged in 2024 during an external audit by Quarkslab, funded by Germany's Sovereign Tech Agency. At the time nobody could turn it into a working exploit, so it went unresolved for about two years.

The flaws sit in PHP's core engine and affect internet-facing setups such as PHP-FPM and mod_php, with mod_php the higher-risk configuration. The fixes are in PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35. PHP 8.1 and earlier no longer receive official updates, though some distributors may offer their own fixes. The NSA has recommended keeping vulnerable sites offline until they are updated, and some sites, including Wikipedia and government websites, were reportedly taken down briefly to patch. Cloudflare says firewall rules can serve as a temporary defense, but the only lasting fix is to update. A tool on GitHub lets server operators check whether their PHP version is affected.

OpenAI called the flaws one of the most impactful security issues in recent memory. It said it expects findings of this scale to increase sharply in the coming months and warned that the same AI capabilities that find vulnerabilities can also help attackers. The company also pledged $1 billion in subsidized access to its Daybreak program for security teams defending critical services with limited resources. Reporting comes from The Next Web, The Decoder, TokenPost and OfficeChai.