shiichan

Patch the Planet Begins: A Five-Stage Global Response to Critical PHP Vulnerabilities

Hi everyone, Shii-chan here! Today's story is a big one about security: how the world responded to critical vulnerabilities that AI helped uncover.

OpenAI News openai.com

What was announced?

OpenAI News introduced Patch the Planet and a coordinated response to critical PHP vulnerabilities. Patch the Planet connects security researchers, maintainers, and incident responders to help address critical vulnerabilities across open source.

For PHP, OpenAI led a coalition of maintainers, researchers, and security providers over roughly two months.

  • Calif found one flaw using GPT-5.6 Sol as part of Patch the Planet
  • Wiz independently discovered and reported both vulnerabilities
  • PHP's maintainers led the fixes and release
  • Wiz, Cloudflare, and others contributed research, technical support, and coordination so protections could be ready before disclosure

Why it matters

PHP is widely used to run websites and web applications, so the impact is broad. The article says frontier AI likely found these flaws before attackers did, creating a "Defender's Window" for preparing fixes. It also warns that the same AI capabilities could help attackers build exploits quickly.

What changes

For operators and administrators, the following tools and support are being offered:

  • A PHP version checker to help identify potentially vulnerable versions
  • A forthcoming Codex skill to help determine which updates apply
  • A $1 billion commitment to subsidized Daybreak access through Daybreak for Frontline Defenders, for organizations with limited security resources
  • Codex for OSS subscriptions for all maintainers, and direct sponsorship of projects such as the Rust Foundation
  • Trail of Bits support for open-source projects (application required)
  • The Patch the Planet website, with disclosures and guidance on finding affected systems, applying updates, verifying fixes, and interim mitigations

Dive Deep

The response is organized into five stages:

  • 1. Validate and fix the upstream vulnerability, including adversarial testing with frontier models to find bypasses and variants
  • 2. Give affected organizations time to act, by deciding who gets advance information so they can prepare protections such as WAF rules
  • 3. Coordinate with government partners: for PHP this included the White House's Super Intelligence Force, the Office of the National Cyber Director, CISA, and international partners
  • 4. Prepare updates across affected products, involving distributors, hosting providers, and response teams
  • 5. Publish guidance and support deployment once a patch is available, and track operator needs

The framework weighs an organization's protective role, its ability to handle sensitive information, and the time it needs to act. Timelines can be reassessed if there is active exploitation, a leak, or a revealing public patch. CISA's BOD 26-04 guidance helps operators decide how urgently to act.

The article also lists four directions for building more secure software:

  • Secure code from the start (safer defaults, catching risky patterns)
  • Migrating critical components to memory-safe languages, with AI assisting and experts verifying compatibility
  • Formal verification, using mathematical proofs of security properties
  • Stronger isolation, sandboxing software that processes untrusted files

OpenAI is also asking the security community for feedback on:

  • Disclosure windows, and what should trigger earlier disclosure
  • When, if ever, governments or operators should emergency-patch systems they manage, and with what safeguards
  • Who needs early notice, and how to weigh coverage against leak risk
  • Secure-by-default packages and backporting fixes to unsupported versions

OpenAI is also calibrating blocking classifiers for Daybreak to reduce misuse without blocking legitimate security work.

Wrap-up

  • Patch the Planet coordinates global responses to critical open-source vulnerabilities
  • The PHP response ran about two months and is organized into a five-stage framework
  • Tools and support include a version checker, a Codex skill, $1 billion in subsidized access, and OSS support
  • OpenAI is asking for community input on disclosure windows and emergency patching

If you run PHP or work on open-source security response, this one's for you!