PHP maintainers blame Chainguard for breaking embargo on critical AI-discovered flaws

Key points

• PHP maintainers released emergency patches for two critical memory-safety vulnerabilities, CVE-2026-109618 and CVE-2026-109619, on October 21, 2026.
• The vulnerabilities, collectively dubbed 'PHP2Shell,' allow unauthenticated remote code execution and affect nearly all PHP versions released in the last decade, including those powering 70% of the internet.
• Researchers at Calif, OpenAI, and Wiz used AI models to discover and exploit the flaws, which had existed in the codebase for over a decade.
• PHP maintainers accused Chainguard of breaking the coordinated disclosure embargo by publishing fixed images six hours before the official release, allowing attackers to study the fix in binary form.
• Chainguard CEO Dan Lorenc defended the decision, stating that evidence of active exploitation in the wild justified shipping the fixes early to protect users.
• The NSA and other agencies urged organizations to immediately patch or take affected systems offline, warning that the vulnerabilities could allow attackers to steal data or deface websites.

Background

This incident follows a broader trend of AI-driven vulnerability discovery, as noted in our September 27 archive on 'LLM-jacking' and AI infrastructure targeting. The PHP flaws represent a significant shift in the cybersecurity landscape, where AI models can rapidly identify and exploit complex, long-standing vulnerabilities that were previously considered impractical to exploit. The coordinated disclosure process, traditionally a cornerstone of responsible vulnerability management, is under strain as AI accelerates the timeline from discovery to exploitation.

How outlets are covering it

The Register and PHP maintainers emphasize the breach of trust and the chaos caused by Chainguard's early disclosure, arguing it undermined weeks of coordinated effort. Wiz and other security firms highlight the role of AI in uncovering these deep-seated flaws, framing the incident as an 'AI inflection point' in cybersecurity. Chainguard, through its CEO, argues that the decision to release patches early was a necessary response to active exploitation, prioritizing user protection over strict adherence to the embargo. The NSA and other government agencies focus on the immediate risk to critical infrastructure and the need for rapid patching or system isolation.

Why it matters

The PHP2Shell vulnerabilities expose a critical weakness in the global web infrastructure, affecting a vast majority of websites. The incident underscores the changing dynamics of vulnerability disclosure in the AI era, where the speed of AI-driven exploitation can outpace traditional coordinated disclosure processes. It raises questions about the balance between protecting users and maintaining the integrity of the disclosure process, as well as the potential for AI to lower the barrier for both defenders and attackers.

What to watch

Organizations are expected to continue patching their PHP systems in the coming weeks, with a focus on identifying and mitigating any systems that cannot be immediately updated. Security firms will likely monitor for signs of exploitation and provide updated guidance. The incident may lead to a re-evaluation of coordinated disclosure policies in light of AI's impact on vulnerability discovery and exploitation.