OpenAI’s Patch the Planet program co-discovers critical PHP engine vulnerabilities affecting 70% of the web

OpenAI’s Patch the Planet program and Wiz independently discovered two PHP engine memory-corruption vulnerabilities—dubbed PHP2Shell—affecting core PHP, which underpins an estimated 70% of the web. OpenAI stated its researchers, including those in the Patch the Planet initiative, identified the flaws using GPT-5.6 Sol, while Wiz reported using Gemini 4 Argon alongside proprietary tooling to uncover the same vulnerabilities.

The vulnerabilities, tracked as CVE-2026-109618 and CVE-2026-109619, stem from PHP’s internal handling of out-of-memory conditions and fatal errors. Researchers demonstrated pre-authentication remote code execution (RCE) against major content management systems, including WordPress, Drupal, and Joomla, as well as potential "Heartbleed-style" memory disclosure across PHP applications. OpenAI and secondary reporting emphasized that exploit demonstrations were researcher-reported and not universally applicable to all deployments.

A two-month coordinated remediation effort followed the discovery, involving PHP maintainers, security providers, and government agencies. OpenAI described the process as a coalition preparing fixes and protections, with confidential advance notifications to governments and external agencies. Eric Wallace, an OpenAI security researcher, and Greg Brockman both reiterated the two-month coordination timeline, though these accounts represent participant perspectives rather than an independently audited timeline.

Cloudflare deployed initial web application firewall (WAF) mitigations ahead of the scheduled patch release, covering known exploit vectors and making them available to all customers, including free-tier users. Cloudflare’s CTO characterized the rules as a stopgap and urged operators to patch at 14:00 UTC on the release date. The NSA advised organizations to prepare for immediate updates or take systems offline if timely mitigation was not feasible.

The incident underscores the role of AI-assisted tools in accelerating vulnerability discovery while also highlighting the complexities of coordinated disclosure. OpenAI framed the effort under its Patch the Planet program, which was introduced alongside a $1 billion subsidy for access to Daybreak cybersecurity models and products. The disclosure has sparked debate about the severity of the vulnerabilities relative to past incidents like Log4Shell and Heartbleed, as well as the broader implications of AI-driven security research.