Over the past two months a coalition of maintainers, researchers and security providers worked to identify, remediate and prepare protections for critical vulnerabilities in PHP. Calif, using GPT‑5.6 Sol as part of the Patch the Planet initiative, identified a critical flaw; Wiz independently discovered and reported both vulnerabilities. PHP’s maintainers led the fixes and releases, with partners including Wiz and Cloudflare contributing research, technical support and coordination to prepare mitigations ahead of disclosure.
What happened and why it matters
Because of the severity of the flaws and PHP’s broad deployment, the incident is described as one of the most impactful security issues in recent memory. The organizations involved say frontier AI found the vulnerabilities before known attackers, creating a “Defender’s Window” that allowed maintainers and security providers time to prepare fixes and protections prior to public disclosure.
The report stresses that the same AI capabilities that help defenders find bugs can also accelerate attackers’ development of offensive tools and exploits, so careful risk management and coordinated disclosure are essential.
Immediate actions for PHP users
Applying the published fixes is the priority. Operators should follow PHP maintainers’ official patch and upgrade instructions and guidance from their providers. The Calif/Codex team published a PHP version checker as a Codex skill to help operators identify potentially vulnerable PHP versions and determine which updates apply.
Operators are also advised to use CISA’s guidance on prioritization and remediation to determine whether emergency patching, immediate mitigations or planned updates are appropriate, and to perform forensic triage where warranted to determine if systems have already been compromised.
A proposed five‑stage response framework
To make responses repeatable and scalable, Patch the Planet proposes a five‑stage framework that was applied in the PHP case:
- Validate and fix the upstream vulnerability: maintainers, researchers and essential technical partners confirm the finding, assess impact and develop and test a fix, including adversarial testing with frontier models to search for bypasses and variants.
- Give affected organizations time to act: maintainers decide who should receive advance information to prepare protections; outreach, information‑sharing agreements and coordination are used so distributors, hosting providers and security vendors can prepare mitigations, backports or WAF rules.
- Coordinate with government partners: government cyber agencies play a role in coordinating a broader response and ensuring guidance reaches organizations responsible for critical infrastructure. In the PHP case, the teams engaged multiple parts of the U.S. government, including the White House’s Super Intelligence Force, the Office of the National Cyber Director (ONCD), and the Cybersecurity and Infrastructure Security Agency (CISA), and worked with international partners.
- Prepare updates across affected products: as fixes, guidance and disclosure plans stabilize, maintainers involve additional distributors, hosting and service providers, response teams and operators to prepare updates and address gaps in coverage.
- Publish guidance and support deployment: publish public guidance so operators can assess exposure, find authoritative updates or interim mitigations, and verify that a fix is running; continue to support operators as they deploy updates and report where they need help.
The framework considers each organization’s protective role, the time they need to act, and their ability to handle sensitive information; maintainers decide whom to involve in advance of disclosure.
Information sharing and risk trade‑offs
Participants receive the information necessary to carry out protective work, from network logs and indicators of compromise to more sensitive working proof‑of‑concept exploits. Because even limited details can reveal a vulnerability, organizers weigh the risk of sharing those details against the protection their sharing enables.
Teams using the framework will reassess timelines if active exploitation, a leak, a revealing public patch, or stalled remediation increases risk to users.
Patch the Planet’s role and additional support
Through Patch the Planet, organizers contributed security expertise and model access for the PHP response; they worked with maintainers and research partners to test fixes and prepare guidance, coordinated with infrastructure providers on protections, and engaged government cyber agencies so guidance would reach critical organizations.
They also committed $1 billion in subsidized Daybreak access through the Daybreak for Frontline Defenders program to support organizations with limited security resources that protect essential services.
Longer‑term goals: prevention and building more secure software
The report emphasizes that AI can assist not only in finding and fixing individual vulnerabilities but also in helping engineers build more secure software from the start. Suggested areas of focus include:
- Secure‑by‑default software packages and better upstream/downstream coordination to enable timely fixes and backports.
- Migrating critical components to memory‑safe languages, with AI assisting migrations, tests and expert review.
- Applying formal verification to critical components to prove specified security properties under stated assumptions.
- Strengthening isolation and sandboxing techniques for software that handles untrusted files or media, with AI used to test and improve sandboxing technologies.
Open questions and community input
The organizers invite feedback from the security community on the proposed approach, highlighting questions such as:
- How much time do maintainers, operators and defenders need to prepare and deploy reliable fixes?
- What evidence should trigger earlier disclosure, particularly when some users can be protected sooner than others?
- Under what circumstances should governments or authorized operators apply patches to systems they manage after losing administrative access, and what safeguards should govern that practice?
- How can downstream packages be reliably fixed or backported when upstream fixes are pending, and what testing and maintenance would make those packages trustworthy options when timely upgrades aren’t feasible?
The initiative also requests recommendations for independent security experts who can review and challenge the framework’s assumptions and identify gaps.
Closing
The PHP case illustrates how frontier AI models can find internet‑scale vulnerabilities and how an effective response requires rapid, coordinated action among maintainers, researchers, providers and government partners. The proposed framework aims to provide a repeatable structure for these events while balancing access to capable models for defenders and protections against misuse.
