Pulling public portals offline isn't usually a sign of a breach. Sometimes it's a calculated panic button.
On October 21, 2026, public sector teams across Europe started yanking government websites off the internet over two critical PHP vulnerabilities—dubbed PHP2Shell—tracking as CVE-2026-109618 and CVE-2026-109619. Both carry a 9.2 severity rating under CVSS v4.0 and can open the door to unauthenticated remote code execution. When a flaw sits inside the runtime engine that powers roughly three-quarters of websites with a known programming language, waiting around for a smooth maintenance window gets real risky, fast. For a deeper dive into this area, we recommend: this related article.
Administrators didn't shut portals down because attackers had already seized control of every server. They shut them down because the timing was awful, the blast radius was huge, and the patch wasn't ready until later that afternoon.
What Actually Happened When Portals Went Offline
There was no single EU-wide order forcing countries to pull the plug. Instead, advisories from bodies like CERT-EU (Security Advisory 2026-018) and the UK National Cyber Security Centre pointed teams toward a stark choice: patch immediately, or isolate exposed systems until you can. For broader information on the matter, extensive reporting can be read at CNET.
In the Netherlands, the national cyber agency took its own website offline, and other high-profile organizations followed, including the Overheid.nl government portal, the Port of Rotterdam, hospitals, major political party sites, and the KNVB football association. Finland's state IT provider Valtori suspended public online services for government agencies as a precaution. France's Interior Ministry pulled prefecture websites, Belgium's foreign ministry took sites down, and Denmark's Kombit suspended services including national school communications. Germany's federal IT provider ITZBund also took several websites offline preventively.
Outside government, even big public-facing web staples felt the squeeze. The Wikimedia Foundation took down affected PHP-backed systems until they could be safely patched, and the Internet Archive went offline for emergency maintenance and security updates.
A precautionary shutdown feels dramatic to citizens trying to renew a permit or check a schedule, yet from an incident-response chair it's often the cleanest way to stop unauthenticated traffic while teams verify builds.
Notice an important detail that early headlines blurred: taking a site down over hacking fears isn't the same thing as confirming a successful compromise. Early on, agencies like Finland's Valtori and Belgium's cyber center reported seeing no signs of breach on their own systems, even as Cloudflare flagged early attack traffic and Amazon Threat Intelligence noted exploitation attempts starting on October 17, four days before the public fix. Later in October, CISA added both CVEs to its Known Exploited Vulnerabilities (KEV) catalog.
Why European Government Websites Were So Exposed to PHP2Shell
People assume government websites run on exotic, custom-built stacks locked behind military-grade walls. Honestly, a lot of public-facing web infrastructure runs on the exact same open-source plumbing as millions of commercial sites—and PHP is everywhere.
W3Techs data cited in coverage around the incident puts PHP around three-quarters of websites with an identifiable server-side language, with WordPress alone powering roughly 43% of all websites. Add to that a warning from the Dutch NCSC (NCSC-NL) that PHP gets embedded inside appliances and management interfaces where local admins don't always get clean, direct package updates.
Memory Errors Before Your App Code Even Runs
The technical teeth of PHP2Shell sit in memory management and error handling—code paths nearly every web request touches. According to The PHP Foundation's technical write-up and threat research from Tenable and Unit 42:
- CVE-2026-109618 involves out-of-memory (OOM) handling, where hitting
memory_limitat an unfortunate moment can leave partially initialized request data that gets treated as valid and corrupts memory. - CVE-2026-109619 involves fatal error handling (
longjmp()) leaving global state pointing to freed memory from an earlier request, creating use-after-free risks across worker lifecycles.
What makes that nasty is the attack surface. On setups like Apache with mod_php, an unauthenticated request can trigger trouble before application code even runs—meaning a hardened CMS or a login wall doesn't magically save you. Other deployment modes like PHP-FPM, php-cgi, FrankenPHP, and LiteSpeed also appeared in advisories depending on execution paths.
The Patch Clock and Early Leak Friction
Patching sounds easy until you're looking at a live production fleet with no fix shipped yet. On October 21, 2026, PHP released fixes across supported branches—8.5.12, 8.4.27, 8.3.36, and 8.2.35—alongside minimal source backports for older branches from 7.4 through 8.1.
Things got even messier around release timing. Security reporting noted that Chainguard published built container images containing the fix at 07:41 UTC on October 21, more than six hours ahead of the coordinated PHP release, drawing criticism from the foundation because early artifacts can give attackers a head start analyzing diffs. Meanwhile, The PHP Foundation openly acknowledged that fixing the OOM flaw had been stalled and restarted over months due to consensus decision-making, limited core maintainers with deep engine expertise, and tough trade-offs around backwards compatibility and third-party extensions.
That candor matters. Open-source security isn't magic; it's human bandwidth and trade-offs.
AI Speed vs. Public Sector Patch Cycles
Another big thread in the coverage was artificial intelligence. Researchers noted that AI tools are helping both defenders find bugs in code and attackers automate scanning and exploitation at machine speed.
From hands-on incident work, here is what I see trip organizations up every single time during a high-severity runtime bug:
- Assuming
php -vtells the whole story. Your CLI binary might be updated while the web SAPI or FPM pool runs a different build. - Forgetting to restart workers. Installing a package without restarting PHP-FPM or web server workers leaves vulnerable processes humming in memory.
- Treating a WAF rule like a permanent cure. Firewall rules buy time, especially when upstream patches haven't landed yet, but they aren't a substitute for updating the runtime.
- Skipping post-incident checks before bringing portals back. If an unpatched system sat exposed during active scanning, flipping the switch back on without checking logs or processes is a gamble.
Regulation is pushing harder on speed too. Under the EU's Cyber Resilience Act (Regulation (EU) 2024/2847), manufacturers of hardware and software products with digital elements have faced phased reporting obligations starting September 11, 2026, for actively exploited vulnerabilities, ahead of broader main requirements in December 2027.
Practical Next Steps If You Run Web Infrastructure
If you manage public-facing websites, internal apps, or appliances, don't treat a scare like this as "someone else's government problem." Run a tight check today:
- Inventory every PHP runtime and embedded appliance, not just your main web servers. Check what NCSC-NL and vendor advisories flag for third-party software bundling PHP.
- Upgrade to a patched supported version (
8.5.12,8.4.27,8.3.36, or8.2.35or newer), and plan migrations off branches nearing end of life—PHP 8.2 hits end of life on December 31, 2026, and PHP 8.3 on December 31, 2027. - Restart all active workers and PHP services after updating, then verify the running version serving actual HTTP traffic.
- Audit logs for anomalous child processes or shells spawned by web or PHP workers if your exposure window overlapped with active exploitation.
When downtime hurts public trust, a rehearsed "isolate early, patch cleanly, verify before reopening" playbook beats improvising at noon on a Wednesday every single time.