Why the websites disappeared
The striking part of PHP2Shell was visible before most people understood the bug: websites disappeared because their operators chose to take them offline. In Parliament on 22 October 2026, the UK government reported 32 unavailable services and four partially unavailable services. The minister said the disruption was largely precautionary while teams applied updates and mitigations [3].
The same parliamentary statement said there was no known confirmed compromise of central-government systems from these vulnerabilities at that time. That is a dated assessment of those systems, not a claim that nobody anywhere was exploiting PHP2Shell [3].
The NCSC’s sequence matters. Identify PHP systems, prioritise exposed services and update. If an update is unavailable, assess the vendor’s temporary mitigations. If those cannot be applied, or take too long, consider taking the affected system offline until it is protected. This was conditional operational advice, not an instruction to switch off every website regardless of its software or exposure [2].
For a service owner, that creates a difficult choice between an outage now and the possibility of an intrusion. The useful preparation is to agree who can make that decision, what customers need to know, and what evidence is required before the service returns. Those decisions are slower when ownership is unclear.
This is the written companion to Bradley’s video, based on the maintainers’ advisories, published research and official statements. The Bitcoin experiment belongs to its organiser. We are not presenting a new Invicta exploitation test or claiming access to an affected organisation’s systems.
The problem sits underneath the website
PHP is the language and runtime that turns application code into a response on the server. A content-management system or framework can run above it. Updating a plugin does not establish that the underlying PHP runtime is fixed. PHP’s advisory covers web-serving installations including PHP-FPM, Apache with mod_php, php-cgi, FrankenPHP and LiteSpeed [1].
| Flaw | What goes wrong | Why it matters |
|---|---|---|
| CVE-2026-109618: out-of-memory corruption | A request stops while internal data is only partly initialised. Cleanup later trusts that invalid data. | A memory-management failure can become attacker-controlled execution. |
| CVE-2026-109619: stale global state | A worker retains a reference to request memory after that memory has been released. A later request reaches it. | Data from different requests crosses a boundary that should have been reset. |
These were not bugs that appeared overnight. The PHP Foundation traces the first signs of the out-of-memory issue to work following a 2024 audit. Fixes stalled because the team could not settle difficult compatibility and performance trade-offs without a demonstrated execution path. The Foundation described that delay as a structural failure. Reports from Calif, OpenAI and Wiz in 2026 established the impact and renewed the urgency [4].
The Foundation says the CVE identifiers circulated on 20 October before the planned coordinated release. PHP moved the release forward from 22 to 21 October while distributors prepared packages. That disrupted sequence helps explain the urgency around the announcement. It does not mean the public advisory permanently lacked a fix: the fixed releases were published on 21 October [4][1].
Attacker: why memory handling can become code execution
In the first flaw, PHP abandons a request after an allocation failure. If that happens at an unsafe moment, cleanup encounters internal structures that were never fully prepared. The PHP team describes attacker-influenced allocations reaching that path. Apache’s mod_php was especially exposed because shutdown could process the invalid state even when request startup failed, before the application ran [5].
The second flaw involves persistent workers. A worker handles one request, releases its memory, then accepts another request. Some error, session or extension state could still point into the old request’s memory. An operation in the later request might use that reference before normal initialisation replaced it. This is a use-after-free: the program continues to trust a reference after the memory’s original lifetime has ended [6].
A vulnerable runtime does not imply that one exploit works unchanged everywhere. Tenable’s research found that reliable execution generally required matching the target’s PHP build and environment. It also warned that appliances sharing one build change that calculation: work done for one configuration may carry over to many matching devices [12].
An exploit that fails against one server does not prove the server is patched. It may simply be the wrong exploit for that configuration. Use package and runtime evidence to establish the fix, rather than treating a failed attack demonstration as a clean bill of health [12].
What the Bitcoin challenge actually proved
James Pavur put a Bitcoin reward behind a deliberately exposed PHP server to test whether the vulnerability could be exploited over the internet. His later report confirms that a participant extracted the challenge data using PHP2Shell. Only one participant successfully extracted useful content; two source IPs triggered the protected helper [7].
The traffic figure needs its caveat: more than 13 million HTTP requests were recorded during about ten hours of reliable web-request logging, not a complete record of the whole challenge. Pavur added a PHP version page at 12:59 UTC on 25 October; the successful helper execution was logged at 15:47. He also recorded PHP crashes before that page existed, so the earlier configuration should not simply be labelled impossible to exploit [7].
The wallet-key mistake meant Pavur verified the result and transferred the reward himself. The important security result was the confirmed extraction from his server, not whether the winner could immediately spend the wallet balance [7].
One successful challenge demonstrates a possible outcome under that setup. It cannot tell a defender how often unrelated systems were breached, or how quickly a particular criminal group could adapt an exploit. The useful question for your own service remains whether its vulnerable code is still reachable.
Defender: investigate the exposure window
AWS reported a PHP2Shell exploitation attempt in its MadPot sensors on 17 October, before public disclosure. Its report distinguishes attempts from proven compromise: AWS said it was unclear whether the observed attempts would have succeeded on real systems. That evidence supports looking back before the public advisory, rather than assuming exposure began when the patch was announced [8].
- Correlate web requests, PHP runtime diagnostics, process events and file changes. An unusual request on its own does not establish execution [8].
- Review unexpected child processes from PHP or the web service, and new or changed files in application directories [8].
- Treat upload-progress requests as investigation leads in context. AWS specifically names PHP_SESSION_UPLOAD_PROGRESS, but a matching field is not a complete intrusion verdict [8].
- Record which systems were exposed, their software state, available log coverage and the time protection became effective. Missing logs are a visibility gap, not evidence that nothing happened.
Updating PHP removes a vulnerability; it does not by itself remove a backdoor already installed through it. The NCSC explicitly calls for assessing possible compromise and checking for backdoors where immediate updating or disabling is not possible [2]. Preserve relevant evidence and involve your incident-response team when the events suggest execution or persistence.
The checks to make on your own systems
Find the runtime that actually serves the site
Ask the hosting owner which PHP installation, worker pool or container handles web requests. Include products that embed PHP; the Dutch NCSC maintains vendor and product status, so use the current list instead of an old product-count headline [10].
Verify the fixed package and replace running workers
Check the installed package against the vendor’s advisory, then confirm the service has replaced its running workers. CERT-EU’s guidance explicitly requires worker restarts for relevant configuration mitigations. Linux vendors may backport fixes, so an unchanged-looking upstream version number is not the whole answer [9].
Check for activity that predates the fix
Review the period the system was exposed, including available logs before public disclosure. Keep the patch record and the investigation result separate: one demonstrates a software change, the other addresses whether an attacker had already acted [8].
These read-only PHP commands report the command-line version and its configuration file locations [11]. Run them only on systems you administer. They are a starting point for inventory, not proof that a different PHP-FPM pool, Apache module or container is running the same build.
php -v
php --iniCERT-EU says the configuration workarounds reduce particular attack paths but do not make an installation safe. Some settings must take effect before request parsing, so changing them inside application code is too late. Have the service owner apply the current vendor guidance and verify it before reopening an isolated service [9].
For managed hosting, ask for evidence that the web-serving PHP package is fixed and the relevant workers have been replaced. A response saying that WordPress and its plugins are current answers a different question. For an appliance, ask its vendor about the product’s own update path [10].
Keep the response practical
Our companion wiki explains the affected technology and includes the hosting-provider email and hunting queries mentioned in the video. Adapt those queries to your own telemetry and check the primary advisories for current updates.
Open the PHP2Shell wikiThe lasting lesson is operational. Know which team owns the runtime, how it is updated, where its logs go and who can isolate it. Then a new advisory becomes a set of decisions about known systems, with evidence for each decision, instead of a search for who might be responsible.
References
Numbered sources support the practical points in this article. Video timestamps will appear when the related media is published.
- Security advisory: Critical remote code execution vulnerabilities, update nowhttps://www.php.net/archive/2026.php#2026-10-21-5
21 October 2026. Fixed releases, affected web deployments and the limits of temporary mitigations.
- Critical security issue affecting PHPhttps://www.ncsc.gov.uk/news/critical-security-issue-affecting-php
21 October 2026. Prioritised actions: discover, update, mitigate, consider isolation and investigate.
- Cyber Security: PHP Vulnerabilityhttps://hansard.parliament.uk/Lords/2026-10-22/debates/14CB4C6D-4E16-4377-8FCF-C2CF77BB5CC0/details
22 October 2026. The minister’s contemporary account of unavailable services and the absence of confirmed central-government compromise.
- Today’s PHP Security Release, and How We’re Strengthening PHP Securityhttps://thephp.foundation/blog/2026/10/21/todays-php-security-release-and-how-were-strengthening-php-security/
21 October 2026. The audit history, stalled fixes, renewed research and disrupted disclosure.
- RCE via OOM corruptionhttps://github.com/php/php-src/security/advisories/GHSA-w8hw-w9fx-j79w
CVE-2026-109618. The request-startup, allocation-failure and cleanup paths.
- RCE via stale global statehttps://github.com/php/php-src/security/advisories/GHSA-6wmf-4rv5-p3rp
CVE-2026-109619. How request-owned memory can remain reachable from a persistent worker.
- Observations from Running stealmybitcoin.comhttps://pavursec.com/blog/stealmybitcoin/
30 October 2026. The organiser’s results and timing; request counts cover only the period with reliable web logs.
- Amazon Threat Intelligence identifies PHP2Shell exploitation prior to public disclosure in MadPothttps://aws.amazon.com/blogs/security/amazon-threat-intelligence-identifies-php2shell-exploitation-prior-to-public-disclosure-in-madpot/
21 October 2026. Sensor observations, their limitations and investigation recommendations.
- Critical Vulnerabilities in PHP, advisory 2026-018https://cert.europa.eu/publications/security-advisories/2026-018/
Updated 23 October 2026. Distribution backports, worker restarts, temporary settings and the separate upload-progress issue.
- Operational information regarding the PHP2Shell vulnerabilitieshttps://github.com/NCSC-NL/php2shell
Maintained product inventory, vendor status and detection resources. Product totals change over time.
- PHP command line optionshttps://www.php.net/manual/en/features.commandline.options.php
The version and configuration commands below describe the command-line installation, not every web worker.
- PHP2Shell one week later: Why one exploit doesn’t work everywherehttps://www.tenable.com/blog/php2shell-one-week-later-why-one-exploit-doesnt-work-everywhere
29 October 2026. Research on exploit portability and the different risk for devices sharing the same build.