PHP2Shell: what happened, and what it means for the systems you own
Two critical flaws in PHP, found with AI, took government sites offline this week. What happened, and the three questions every leader should be able to answer.
What's going on
This week, the PHP project shipped emergency fixes for two critical flaws in the PHP engine itself, nicknamed PHP2Shell (CVE-2026-109618 and CVE-2026-109619, both rated 9.2 out of 10). PHP sits behind roughly 70% of websites whose server-side language can be identified, including WordPress, Drupal and Joomla. The flaws let an attacker run their own code on a server, or read sensitive memory, with nothing more than a crafted web request. No login needed. They were found by security researchers using frontier AI models, after going unnoticed for more than a decade. Details leaked a day early, governments in Australia, Europe and the US took sites offline as a precaution while they patched, and attackers are now actively trying to exploit unpatched servers.
How the flaw works, in plain terms
When a web request ends badly, say it runs out of memory or hits a fatal error, PHP can leave a little mess behind in its memory. A carefully built request, often one that looks like a file upload, can create that mess on purpose. The next request handled by the same server process then trusts the leftovers. That's enough to take over the process, or to read data belonging to other visitors: passwords, tokens, API keys.
The important part for leaders: the flaw sits in PHP, not in your website's code. A perfectly built WordPress site on an unpatched server is still exposed. Plugins can't fix it. The only complete fix is updating PHP on the server to 8.5.12, 8.4.27, 8.3.36 or 8.2.35, or a patched package from your hosting vendor. Firewall rules from providers such as Cloudflare help as a stopgap, but the PHP team is clear that updating is what actually fixes it.
Who's been affected
In Australia, the ASD's own cyber.gov.au had an outage, and sites including the ABS and Jobs and Skills Australia were briefly unavailable while work was done to mitigate the risk. GovCMS took its platform off direct internet access overnight. Finland took its government public web services offline. The Dutch cyber security agency took its own site down. The UK, Japan and the US issued urgent guidance.
These were precautionary takedowns. The ACSC says it is not aware of successful exploitation in Australia, and no government has reported a breach from this flaw. But exploitation is real: AWS logged an attempt four days before the public disclosure, and says the bug is now being broadly exploited in the wild.
The AI angle
This is the part leaders should sit with. The researchers at Wiz say the underlying bugs had been there for years, but proving they could be exploited required a chain of steps "at the edge of what human researchers could practically achieve." Frontier AI models made it practical, turning that work into hours.
The same capability cuts both ways. AWS put it simply: AI makes it cheaper to turn a published fix into a working attack, so more people can act within hours of a disclosure. A 30-day patch window used to be normal. For internet-facing systems, the guidance now is to measure it in hours.
We don't see that as a reason for alarm. We see it as a reason to know your systems.
Why this matters for leaders
Most mid-market businesses don't know, off the top of their head, what their website, customer portal or internal tools run on. That's fair. But this week, the organisations that moved fastest were the ones that could answer three questions in minutes:
- What do we run? Every site, portal and tool, including the ones a vendor hosts for us.
- Who patches it, and how fast? A name, not a department.
- What data sits behind it? Because memory leaks expose whatever the server touches.
If those answers take a week of emails, that's the real risk. Not PHP.
Built right, and owned
This is where we think the conversation goes next. A lot of business software is rented, stitched together from plugins and maintained by whoever set it up years ago. It works, until a week like this one.
The alternative isn't "never use open source." It's building systems deliberately: a modern stack kept up to date, clear ownership of every component, and someone accountable for updates. When you own the system, you know what's inside it.
The code we write doesn't run on PHP. No stack is immune, though, so the real job is knowing every component you rely on and keeping it updated. That's a choice we make on every build.
That doesn't make anyone immune to the next one. What it does is make the next one a short conversation instead of a scramble.
What to do this week
- Ask your web host or developer: is anything we run on PHP, which version, and is it patched for CVE-2026-109618 and CVE-2026-109619?
- Ask your SaaS vendors for their PHP2Shell statement. Many have published one.
- If something can't be patched today, follow the ACSC's advice and consider taking it offline until it can.
- Ask whoever patched to check logs back to 17 October, not just from the disclosure date.
- Write down who gets the call next time.
The bigger picture
AI will find more of these. That's good news: decade-old flaws are being found and fixed. The businesses that do well will be the ones whose systems are well understood, well maintained and genuinely theirs.
That's the work we do. We build big, complex systems that give businesses new capability, built right and owned by the business. If you'd like to see how we think about it, start at navigatr.ai.
Sources: PHP security advisory, Wiz Research, ASD's ACSC, AWS Security Blog, BBC, ABC, ACS Information Age.