PHP2Shell exposed a hard problem for vulnerability response: a patch can help defenders only if they can get it before attackers use it. The two critical flaws in PHP, a widely used programming language, carry a reported severity score of 9.2. Their disclosure also showed how quickly a coordinated release can unravel once technical clues and exploit attempts begin to circulate.
The dispute centers on when the flaws became exploitable, and whether publishing fixes early helped or hurt. The PHP Foundation says Chainguard broke an agreed embargo by releasing patched images before the scheduled time. Chainguard CEO Dan Lorenc says the company acted after finding evidence of exploitation and judging that details about the vulnerabilities and fixes were already spreading. Cloudflare and AWS have reported signs of attack before the public announcement.
Those competing accounts matter to security teams deciding how much time they can count on between an advisory and active attacks. Frank Dickson of Dickson Research put the dilemma plainly: “Embargoes matter. This one was nearly impossible to keep.”
A disclosure plan that lost its lead time
The flaws, tracked as CVE-2026-109618 and CVE-2026-109619, can enable remote code execution. The Foundation says a contractor first identified signs of the out-of-memory issue behind CVE-2026-109618 in 2024. At the time, researchers had not found a concrete attack path, and other issues with clearer impact took priority.
Work intensified on July 30, 2026, after the PHP project joined OpenAI’s Patch the Planet program. Researchers from Calif, OpenAI, and Google Cloud’s Wiz clarified the issue’s full impact, the Foundation said, putting a fix at the top of the project’s priorities.
Distributors, including Chainguard, received patches ahead of release so they could prepare updates for customers. The planned disclosure time was October 22 at 14:00 UTC. That schedule began to shift on October 20, when the two CVE identifiers appeared online at about 15:00 UTC. The Foundation says technical details had not leaked at that point, but the PHP project moved up its pre-announcement, publishing it at 19:30 UTC, and advanced the patch release to October 21 at 14:00 UTC.
The earlier announcement set off another disclosure. Calif mistakenly posted a video showing a proof-of-concept exploit against multiple targets, along with limited technical information. The video was later removed, but copies had already begun spreading on social media.
Then, at 07:41 UTC on October 21, Chainguard published images containing the fix, around six hours before the revised release time. The Foundation criticized the move, saying it disregarded weeks of coordination among the parties involved. The PHP project released patches publicly at 14:00 UTC, 24 hours ahead of its original schedule.
The Foundation said distributors were still preparing packages, and an earlier public release could have left users without a straightforward update path. It described the one-day change as a compromise: move quickly while giving users and distributors time to prepare.
Attack reports complicate the timeline
Chainguard disputes the Foundation’s account of what justified its decision. Lorenc said the company considered how readily it could independently reproduce the flaw and build an exploit, how much information about the upcoming vulnerabilities and fixes was already available, and reports of attacks in the wild. He set out that reasoning in a LinkedIn post.
Cloudflare CEO Matthew Prince said his company observed “thousands” of PHP2Shell exploitation attempts about 10 hours before the public announcement. According to Prince, activity rose to “tens of thousands” in the 45 minutes after the announcement, reached hundreds of thousands within two hours, and later climbed into the millions.
AWS reported an attempted exploit on October 17, four days before public disclosure. The request came from an IP address linked to a commercial anonymization network, AWS CISO CJ Moses said. AWS characterized the activity as an attempt by a threat actor, while saying it did not know how the actor obtained information about the flaw beforehand. The earlier activity raises a central question in the dispute: whether the embargo still offered defenders meaningful protection by the time the patch was released.
Some organizations responded by taking websites offline, according to the source reporting. The scale of that reaction, alongside the reported volume of attempts, underlines why timing and update access matter well beyond the PHP project itself.
Embargoes need a shorter, tighter window
Dickson argues that reports of leaks and early attacks do not excuse publishing a fix ahead of the agreed release. Distributors received advance access so their users could update when the flaw became public. Releasing the patch early also gave attackers a way to study the change while many PHP users were still waiting for packages.
“An embargo protects defenders only until someone else finds the same bug,” Dickson said. A longer wait can shrink the period in which coordinated access gives defenders an advantage, and each additional recipient increases the chance that information will escape. His prescription is a shorter disclosure window, fewer early recipients, and a shared understanding that the patch itself contains useful clues. “A patch is a treasure map,” he said.
Beauceron Security CEO David Shipley sees the same problem in the mechanics of secrecy: more people with access and more time create more opportunities for a leak. He also argues that AI tools can make technical clues easier to turn into working exploits. With enough information about a flaw, an agent can be directed to inspect software and help work out how it behaves.
Prepare for exploitation before the advisory
Shipley expects PHP2Shell’s effects to persist, comparing its potential spread to Log4j. PHP runs across prominent online services, and vulnerable components can also make their way into connected devices and other products. That makes asset and dependency records practical response tools. A software bill of materials can help teams identify where affected software appears, but maintaining an accurate view of the estate still takes routine work.
Shipley also points to market incentives that reward speed and visible product releases, even when they build security debt. He argues that expectations for product makers and AI tool providers need to reflect the risks their systems introduce. In his view, the problem predates AI: organizations have long faced pressure to ship first and address security later.
For security leaders, Dickson’s advice is operational. Map where each runtime is deployed, including software supplied by vendors. Decide in advance who can take a service offline, so teams can make that call under pressure without negotiating authority during an incident. Build detections around behavior, such as unexpected processes or new files on a web server. A signature tied to a specific CVE may not be available on the first day of an attack.
Dickson’s summary captures the tension: “Chainguard jumped the gun, but the race had already started.” Coordinated disclosure can still give defenders time to prepare, but PHP2Shell shows that teams should plan for exploitation to start before a public advisory. The useful response is to tighten release coordination and make sure asset records, update paths, and behavior-based detections are ready before the next disclosure arrives.
This article originally appeared on InfoWorld.