Millions of attempts to exploit two critical vulnerabilities in PHP were reported in the days after public disclosure, turning a serious software flaw into an urgent patching problem for website operators around the world. The scale of activity shows how quickly technical details can be converted into automated internet traffic, but the number of attempts should not be confused with millions of successful intrusions.
A steep rise in exploit attempts
On 22 October, Cloudflare chief executive Matthew Prince said the company had observed millions of attempts to exploit the flaws, collectively known as PHP2Shell. According to his public account, Cloudflare saw thousands of attempts during the 10 hours before the vulnerabilities were announced, tens of thousands in the first 45 minutes after disclosure, and hundreds of thousands within two hours before the total climbed into the millions.
Cloudflare sits between many websites and the wider internet, providing traffic delivery and security services, so its network can give it broad visibility into malicious requests. The reported volume nevertheless represents observed attempts, not a reliable count of unique victims, successful code execution or confirmed data theft.
The flaws are tracked as CVE-2026-109618 and CVE-2026-109619. Both received critical severity ratings and could enable unauthenticated remote code execution in affected configurations. PHP maintainers released security updates on 21 October for supported branches, including PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35. The PHP project warned that updating was the safe response and that there was no complete workaround.

Activity before disclosure raises questions
The chronology raised concern that information about the vulnerabilities may have circulated before the planned announcement. Cloudflare’s observations began before the public release. Amazon Threat Intelligence later reported that its MadPot honeypot had recorded an exploitation attempt on 17 October, four days before disclosure. Amazon said it remained unclear how the actor had obtained information about the vulnerability while it was still under embargo.
That evidence merits investigation, but it does not identify a source of a leak or prove that every early request was a genuine exploit. Security researchers, automated scanners and threat actors can generate traffic for different reasons. Investigators need to compare payloads, timing, infrastructure and the vulnerable conditions being tested before attributing activity to a particular group.
Several European public-sector sites were taken offline or placed into maintenance as a precaution while their operators assessed exposure. Such decisions can reduce immediate risk, but they do not by themselves demonstrate that a system has been breached.

Attempts are not the same as compromises
The distinction matters because vulnerability reporting can move faster than forensic confirmation. A sensor may count a request that matches an exploit pattern, a probe that tests whether a server is vulnerable, or a payload that fails before execution. None should automatically be reported as a successful compromise.
Security firms published evidence of widespread scanning and exploitation attempts after disclosure. Their assessments of successful exploitation have not all been identical. Sophos reported activity it assessed as one successful exploitation, while Palo Alto Networks’ Unit 42 later said it could not confirm that the initial access in a reported European government incident was linked to these specific PHP2Shell flaws and retracted its earlier attribution.
The public evidence therefore supports urgency, not blanket claims that millions of servers were taken over. The important open questions are how many exposed systems remain unpatched, how many attempts achieved code execution, and whether any confirmed intrusions led to persistence, credential theft or broader compromise.
The patching priority
PHP underpins websites, content-management systems and web applications. Its widespread deployment means an organisation can be exposed through a hosting stack, a control panel or a packaged product even if its own development team did not write the vulnerable runtime code.
The fixes address two critical flaws in PHP request handling. Because attacks may target the runtime before application-level protections can help, a web application firewall or a set of blocked URLs should not be treated as a replacement for installing the relevant security update. Operators using end-of-life PHP branches need to check their vendor’s support and patch guidance rather than assume the newest version number is available to them.

What operators should do now
Administrators should inventory PHP across web servers, containers, control panels and managed hosting accounts; confirm exact runtime versions with the provider; and install the appropriate patched release or vendor backport. Internet-facing systems should be prioritised. After patching, teams should verify that the running service uses the fixed code and review logs for unusual requests, unexpected processes, modified files, new accounts or suspicious outbound connections.
Where signs of compromise exist, patching alone is not enough. The organisation should preserve relevant logs, investigate the system, rotate exposed credentials where indicated and follow its incident-response procedure. If a critical service cannot be updated promptly, restricting public access or temporarily isolating it is safer than assuming that monitoring alone will stop exploitation.
The PHP2Shell episode is a warning about speed and visibility. Cloudflare’s numbers show that the volume of malicious traffic can rise sharply within hours of disclosure; Amazon’s earlier observation shows that suspicious activity can precede an official advisory. For defenders, the durable lesson is to prepare for emergency patching, investigate before assigning blame, and distinguish a flood of attempts from evidence of a confirmed breach.
Forensic Times



































