AI discovers critical PHP flaws, sparking disclosure debate
Two critical PHP vulnerabilities forced emergency patches and triggered a dispute over early release by Chainguard amid AI-driven discovery.
On October 21, 2026, the PHP project released emergency fixes for two critical security flaws, prompting many website operators to take services offline. The incident, dubbed PHP2Shell, involved vulnerabilities CVE-2026-109618 and CVE-2026-109619, which allow remote code execution. While the technical fixes were urgent, the event ignited a significant controversy regarding coordinated vulnerability disclosure in an era where AI accelerates exploit development.
What happened
The vulnerabilities affect PHP versions powering approximately 70% of websites. The PHP Foundation issued fixed releases for branches 8.2.35, 8.3.36, 8.4.27, and 8.5.12. Users on older, unsupported branches must upgrade or rely on distributor backports. WordPress advised site owners to update their PHP installations, clarifying that the flaws reside in the language itself rather than the WordPress application.
The discovery process highlighted the growing role of artificial intelligence in security research. Security firm Calif, participating in OpenAI’s Patch the Planet initiative, used GPT-5.6 Sol to identify a path to remote code execution, reporting it on August 11. OpenAI independently reported the same issue that day, followed by Wiz on August 24. Although PHP maintainers had noticed signs of an underlying out-of-memory problem during a 2024 audit, they lacked a concrete exploitation route until these AI-assisted reports clarified the severity.
A dispute erupted when Chainguard, a software supply chain security company, released patched container images more than six hours before the coordinated disclosure time. Chainguard CEO Dan Lorenc stated he broke the embargo due to reports of active exploitation in the wild and leaking details. The PHP Foundation criticized this unilateral move, arguing it gave attackers time to study the fix while other users remained vulnerable. Lorenc apologized for the stress caused but maintained that protecting users was the priority.
How it works
The core technical issue involves an out-of-memory condition that can be manipulated to achieve remote code execution. This type of vulnerability allows an attacker to run arbitrary commands on a server hosting a PHP application. The flaw is particularly dangerous because of PHP’s ubiquity; as Moshe Siman Tov Bustan of OX Security noted, any publicly accessible PHP instance is potentially exposed.
The controversy centers on the mechanics of coordinated disclosure. Typically, vendors agree to an embargo period to prepare patches simultaneously. However, AI models are now capable of finding and verifying exploits rapidly. Chainguard’s testing harness, using a frontier AI model, independently rediscovered the issue. When Lorenc received reports of potential live exploits and saw proof-of-concept videos leaking online, he decided the risk of waiting outweighed the benefit of coordination. He released the fixes publicly to ensure immediate access, bypassing the standard embargo protocol.
Key details
- The vulnerabilities are identified as CVE-2026-109618 and CVE-2026-109619.
- Fixed PHP versions are 8.2.35, 8.3.36, 8.4.27, and 8.5.12.
- Chainguard released patched images early, citing evidence of active exploitation.
- The PHP Foundation labeled the delay in fixing the 2024-identified issue a "structural failure."
- Amazon Threat Intelligence detected exploit attempts against honeypots starting October 17.
- AI models like GPT-5.6 Sol played a key role in discovering and verifying the flaws.
Why it matters
For software engineers and IT leads, this incident underscores that traditional maintenance windows may no longer suffice for critical infrastructure. The speed at which AI can discover and weaponize vulnerabilities means that the window between discovery and exploitation is shrinking. Organizations must be prepared to deploy patches immediately, even outside normal schedules, to mitigate risks from widely used dependencies like PHP.
The dispute also highlights the fragility of current open-source security models. The PHP Foundation admitted that resource constraints and complex trade-offs delayed a fix for a known issue. As AI lowers the barrier for finding bugs, the burden on maintainers increases. Without additional funding and support from the companies and governments that rely on this software, similar structural failures are likely to recur, leaving the web ecosystem vulnerable to internet-scale threats.
What you can do
- Update your PHP installation to version 8.2.35, 8.3.36, 8.4.27, or 8.5.12 immediately.
- If you use an unsupported PHP branch, plan an upgrade or apply distributor backports.
- Review server logs for suspicious activity dating back to October 17, 2026.
- Assume compromise if you find evidence of exploitation on unpatched systems.
- Engage with trusted suppliers like Red Hat or Canonical for validated patches.
- Advocate for organizational support of critical open-source projects you depend on.



