PHP2Shell Exposes Disclosure Strains and a Longstanding Maintenance Gap

The PHP2Shell emergency put two questions in front of the open-source security community: how quickly should a fix ship when attackers may already be probing a flaw, and why did a weakness first noticed in 2024 take until 2026 to become a concrete remote-code-execution risk? The answers matter well beyond PHP. The language runs a large share of the web, and its response shows how AI-assisted research is tightening the time available to find, fix and distribute security updates.
Operators took some websites offline while waiting for patches for two critical PHP flaws. That precaution doesn't mean those sites were compromised. PHP's reported presence on roughly 70% of websites describes the language's reach, not the number of systems that were vulnerable or attacked. WordPress advised site owners to update PHP; the vulnerabilities don't require a WordPress update.
The affected issues are CVE-2026-109618 and CVE-2026-109619. PHP fixed them in versions 8.2.35, 8.3.36, 8.4.27 and 8.5.12. Operators on older branches need to move to a supported release or obtain a backport from their distributor. For teams running publicly accessible PHP, the immediate task is to deploy an appropriate fix and review logs for suspicious activity.
The release dispute turned on evidence
The patch rollout became contentious after information about the flaws began circulating before the coordinated release on October 21. The PHP Foundation said the two CVE identifiers appeared on October 20, followed by an accidentally published proof-of-concept video. Chainguard CEO and co-founder Dan Lorenc then made patched container images available more than six hours before the agreed release time.
Lorenc said he acted because vendors had reported what they believed were exploits in the wild. In his account, Chainguard learned on Friday, October 16, that a PHP issue was under investigation, independently found the issue that night using its testing harness and a frontier AI model, and received a full briefing the next day. The company also helped test fixes. By Tuesday night, October 20, he said, reports of possible active exploitation had intensified.
Three factors shaped his decision: how readily Chainguard had reproduced the vulnerability and built an exploit, how much information about the flaws and fixes was already public, and the reports of activity against vulnerable systems. Lorenc said he believed coordinated disclosure guidelines allowed an early release when there was evidence of exploitation. Chainguard made the images publicly accessible, rather than limiting them to customers, and notified the disclosure group.
The PHP Foundation disagreed with that reading. In its account of the release, it said, “Chainguard unilaterally broke the embargo and published built images containing the fix.” The Foundation had provided Chainguard with the patch so customers could update at the planned time. Publishing images early, it argued, gave attackers a way to inspect the fix while other PHP users still lacked updates.
The disagreement came down partly to what counted as a working exploit. Lorenc cited reports from other vendors. The Foundation said its security team hadn't found a functional exploit among the proof-of-concept material circulating at release time. Amazon Threat Intelligence later reported an apparent exploit attempt against one of its honeypots on October 17 and more attempts early on October 21, but said it couldn't determine whether those attempts would have worked against real systems.
There are costs on both sides. An embargo gives maintainers and distributors time to prepare fixes for release together. An early patch can help organizations that can obtain it, while also giving attackers clues about the vulnerability before updates reach everyone. A previous WordPress incident illustrates how quickly scanning can follow a public fix: after WordPress released version 7.1.2 on September 22 for CVE-2026-87902, Patchstack observed probes that day and attempts to write files the following day. That episode doesn't establish that PHP2Shell was successfully exploited when Chainguard published its images.
The older issue is the deeper warning
The PHP Foundation acknowledged a failure that predates the disclosure dispute. Developers had seen signs of the out-of-memory problem behind one of the PHP vulnerabilities during a 2024 audit. At the time, they hadn't established a concrete path to remote code execution. Potential fixes brought difficult security and compatibility trade-offs, and work repeatedly stalled.
The Foundation described the delay as “a structural failure” tied to its decision process and lack of resources. That admission shifts the focus from the timing of a single patch release to the conditions that determine whether a known weakness gets resolved. Reporting on paid maintenance has connected dedicated support with stronger security practices. The Foundation is now seeking more help from companies, hosts and governments that rely on PHP.
PHP's maintainers led the fixes after security researchers identified the flaw's full impact in August. Security firm Calif found a route to remote code execution using GPT-5.6 Sol through OpenAI's Patch the Planet initiative and reported it to PHP on August 11. OpenAI independently reported the same underlying issue that day, while Wiz submitted its findings on August 24. The episode adds to the pressure on open-source maintainers as AI-assisted researchers produce findings and develop exploits faster.
Open-source projects are already dealing with a flood of AI-generated vulnerability reports. PHP2Shell adds a harder coordination problem: once technical details start leaking, maintainers must weigh the value of synchronized updates against signs that attackers may be acting before a public fix is available. Discussion among open-source leaders increasingly centers on how to make that response workable under pressure.
Security research is reaching more software
Several industry voices see PHP2Shell as part of a wider change in vulnerability research. Peter Nelson, chief technology officer at CIQ, Rocky Linux's founding sponsor, said AI had been used to exploit kernel vulnerabilities during the prior six months, with attention now spreading to widely used programs such as PHP. He expects open-source projects to use AI and other methods to harden code, while warning that the tools used to find weaknesses may become more accessible.
“Today, such vulnerability exploitation requires a frontier model. Tomorrow that won’t be the case.”
Lorenc called the challenge an industry-wide one, arguing that defenders will need to work together to secure widely used infrastructure quickly enough to limit harm. His account of the early release and the Foundation's criticism show how difficult that coordination can be, even when the participants share the goal of protecting users.
Canonical offered a more favorable assessment of the broader process. Luci Stanescu, its security engineering manager, said the company supports coordinated vulnerability disclosure and considered the third-party embargo leak unfortunate, while judging that the disclosure process itself worked well. That view differs from the Foundation's criticism: a leak can damage part of a coordinated release without making the entire effort unsuccessful.
What operators should do now
For organizations that run affected PHP systems, the operational guidance is direct. Gunnar Hellekson, Red Hat's vice president and general manager for Lightwell, said stable production software can be easy to overlook even as AI-driven research increases the pace and volume of vulnerability discoveries. He urged customers to deploy tested patches from trusted suppliers, including Red Hat, as soon as they become available, even outside normal maintenance windows.
That advice aligns with the immediate risk: teams need to identify their PHP versions, apply a fixed release or distributor backport, and inspect systems for signs of attempted exploitation, including activity dating back to October 17. OX Security's Moshe Siman Tov Bustan advised operators with publicly accessible, unpatched PHP instances to assume they are exposed; if they find evidence of exploitation, he said, they should treat the system as compromised. CEO Neatsun Ziv described the potential reach as internet-wide because PHP runs across many industries and products.
The companies that depend on open-source software also shape the resources available to maintain it. Kara Sprague, CEO of HackerOne, said software supporting millions of websites can't rely on a small group of maintainers to manage increasingly sophisticated threats alone. That point echoes the Foundation's account of a fix delayed by limited resources and hard technical trade-offs.
PHP2Shell's patch is a necessary response, but the longer-term lesson lies in the timeline: researchers found the exploit path in 2026, while maintainers had seen an early sign of the underlying issue in 2024. Faster discovery makes that gap more costly. Better-supported maintenance and disclosure procedures that can withstand leaks and uncertain threat reports will determine how often the next fix reaches users before attackers can act.
The post AI found a critical PHP flaw. The bigger failure began in 2024. appeared first on The New Stack.