How We Responded to a Critical PHP Vulnerability
A critical vulnerability in PHP's request handling prompted a fleet-wide response. Here's what we did and what we changed.
Recently, two critical vulnerabilities in PHP’s request handling were publicly disclosed. The flaws affected how PHP manages memory during request cleanup and how internal state is shared between requests on the same worker. Together, they allowed unauthenticated remote code execution on affected servers. Every major PHP server interface was affected, and any hosting provider or SaaS platform running PHP needed to assess their exposure immediately.
EventWorks runs on PHP. We took notice.
Assessing and responding
Within hours of the disclosure, we audited every server in the fleet. The priority: identify which environments were exposed, patch or isolate them, and confirm no exploitation had occurred.
Where patching in place was fastest, we did that. Where a cleaner path was to reprovision, we migrated applications to freshly built environments on patched runtimes. We tested each migration against production data before switching traffic over.
We reviewed access logs, intrusion detection alerts, and application telemetry across the full fleet. No evidence of exploitation or unauthorised access. No customer data was affected.
What we changed
Fixing the immediate problem was necessary. Making the same class of problem less likely to recur was the real work.
We stopped treating vulnerability management as a periodic exercise and built it into the release process. Automated scanning now runs as part of the deployment pipeline, checking every release against known vulnerability databases before it reaches production. A flagged dependency or runtime component blocks the deployment until it’s resolved.
That covers the application layer. At the infrastructure layer, host-level scanning runs continuously across the fleet, checking installed packages against published advisories whether or not a deployment is happening. Neither check relies on someone remembering to run it.
We also shortened our patch cycle and automated reboots for kernel-level updates, so security patches don’t sit waiting for a maintenance window that never comes.
The full picture
The disclosure was outside our control. How we responded, and what we changed afterwards, was not.
We’ve published a dedicated security page covering how we handle infrastructure security, network isolation, monitoring, backups, and incident response.
We’re also working towards Cyber Essentials certification, which will give us independent verification against a recognised baseline. More on that when it lands.