OpenAI reported the discovery and remediation of two critical PHP vulnerabilities through its Patch the Planet initiative. Patches for the flaws were released on Oct. 21, 2026, as the FBI warned that they could allow malicious actors to remotely execute code in PHP core and potentially compromise sensitive data and networks.

The vulnerabilities, tracked as CVE-2026-109618 and CVE-2026-109619, affect PHP, a widely used programming language that powers websites and web applications. OpenAI said Calif found one of the vulnerabilities using the GPT-5.6 Sol model, while Wiz specialists independently discovered both.

PHP maintainers led the patching effort, with contributions from OpenAI, Wiz and Cloudflare, according to OpenAI.

FBI and OpenAI urge immediate updates

In an Oct. 21 post, the FBI urged organizations to update affected PHP installations immediately and inventory every location where PHP runs, prioritizing internet-facing systems and vendor products that bundle the language. If an immediate update is not feasible, the agency recommended taking the affected system offline until it can be fully mitigated.

The FBI also advised organizations to monitor web servers for suspicious activity, including PHP processes unexpectedly launching command shells or downloading files. Organizations that identify signs of compromise should isolate affected systems, preserve web server and PHP logs, and report the incident to a local FBI field office or through the agency’s online reporting service.

The agency directed organizations to PHP’s announcement, CERT-EU Security Advisory 2026-018 and the U.S. Cybersecurity and Infrastructure Security Agency’s Known Exploited Vulnerabilities catalog for updated information.

OpenAI said installing the patches should be the top priority for organizations using PHP. It also released a PHP version checker as a Codex skill to help administrators identify potentially vulnerable versions and determine which updates they need.

According to a post by CZ, details of the vulnerabilities leaked online prematurely, and exploitation in real-world attacks was confirmed almost immediately after disclosure. CZ said the flaws could threaten servers running widely used platforms, including WordPress and Magento. IT teams began deploying web application firewall rules and installing updates to protect systems.

OpenAI outlines coordinated response

OpenAI called the case one of the largest security incidents in recent times because of the vulnerabilities’ severity and PHP’s widespread use. The company said that, to the best of its knowledge, advanced AI models identified the flaws before malicious actors did, giving developers and security vendors time to prepare patches and mitigations before public disclosure.

OpenAI also warned that attackers could use the same AI capabilities to build attack tools and exploits rapidly.

Based on the PHP response, OpenAI proposed a five-stage approach for handling critical vulnerabilities in widely used software. The first stage brings maintainers, researchers and technical partners together to confirm the vulnerability, assess its impact, develop a fix and test it against potential bypasses.

The second stage provides advance notice to organizations that need preparation time, allowing distributors to adapt fixes for older versions and service providers to test protective measures. The third coordinates with government partners that can distribute guidance to organizations responsible for critical infrastructure and essential services.

The fourth stage involves distributors, hosting providers, service providers and incident-response teams in preparing updates for vulnerable products. The final stage covers publication of update instructions, temporary mitigations and system-verification recommendations, along with support for the rollout.

OpenAI said response timelines should change if there are indications of active exploitation, a data leak occurs or delays in delivering a fix increase risks to users.

Through Patch the Planet, OpenAI also works with government cybersecurity agencies and helps coordinate efforts to protect vulnerable systems. Separately, the company announced plans to provide $1 billion in subsidized access to Daybreak through the Daybreak for Frontline Defenders program, which is intended for organizations that protect essential services but have limited security resources.

Source: Incrypted