OpenAI says its models helped find critical security flaws in PHP

OpenAI announced that its models helped researchers, working under its "Patch the Planet" initiative, discover critical security flaws in PHP, which it says runs an estimated 70% of websites. According to OpenAI, it led a coalition of maintainers, researchers and security providers for about two months to prepare fixes and protections before disclosure; the fix was intricate and required a major change to the PHP engine.

In its full post, OpenAI says Calif identified a critical flaw using GPT-5.6 Sol under Patch the Planet, that Wiz independently found and reported both vulnerabilities, and that PHP's maintainers led the fixes and release. It says it worked with partners such as Wiz and Cloudflare, and engaged U.S. government cyber agencies (including CISA and the Office of the National Cyber Director) and international partners.

OpenAI says every site can now upgrade PHP to stay secure, in addition to applying mitigations from security vendors. It released a PHP version checker on GitHub, with a Codex skill to follow, to help operators spot potentially vulnerable PHP versions. OpenAI also says it has committed $1 billion in subsidized Daybreak access for organizations with limited security resources, and it proposes a five-stage framework for coordinating responses to critical vulnerabilities (validate and fix upstream, give affected organizations time to act, coordinate with governments, prepare updates, then publish guidance) and asks the security community for feedback. PHP's site posted a notice of the changes on October 21. Cloudflare said it made stopgap WAF rules for the flaws available to all customers, including free plans, and, per a Cloudflare write-up, the vulnerability leaked early, so its rules went live about nine hours before disclosure.

OpenAI calls this one of the most impactful vulnerabilities in recent memory and says that, to its knowledge, frontier AI found it ahead of attackers. Those are OpenAI's claims. The announcement had more than 1,000 likes on X within minutes of posting.

Wiz Research's account (PHP2Shell). Wiz, one of the groups OpenAI worked with, published its own write-up the same day. Wiz says:

• The flaws are memory-safety bugs in how PHP handles fatal errors, especially out-of-memory errors, grouped under two CVEs (CVE-2026-109618 and CVE-2026-109619, severity up to CVSS 9.2). It says every PHP version back to 5.4 is affected by at least one of them.
• It chained them into an exploit script, "php2shell", and demonstrated unauthenticated remote code execution on default WordPress, Drupal and Joomla container images, as well as disclosure of PHP worker memory (which can hold passwords, tokens and API keys).
• The team used Gemini 4 Argon and GPT-5.6 Sol throughout, and says Argon contributed most to finding the bug, building the exploit and validating the patch. A Google DeepMind executive is quoted praising that result.
• Timeline: first vulnerability found on August 19, reports sent to the PHP security team on August 24, PHP acknowledged them on September 21, and patched releases and public disclosure came on October 21.

Wiz also thanks Calif and OpenAI for discovery, coordination and patch verification, and lists Cloudflare, AWS, Google, Microsoft, Akamai, Fastly and others as having prepared mitigations. These are Wiz's own claims.