Artificial intelligence has exposed two critical security flaws in PHP, the programming language that underpins much of the web, prompting organisations across Europe to shut down websites as a precaution. The vulnerabilities, which went undetected for more than a decade, were turned into working attacks by AI models from OpenAI and Google, marking a new phase in cybersecurity.
The Dutch government was among the first to react, taking offline its DigiD digital identity system, the Overheid.nl portal, and the website of the lower house of parliament. The Port of Rotterdam, several hospitals, and the country's five largest political parties followed suit. Even the Dutch National Cyber Security Centre pulled its own site. Finland's state IT provider Valtori switched off government web services wholesale, while France's Interior Ministry put local administration sites into maintenance mode. Germany's federal IT service ITZBund took down several portals.
CERT-EU, the cybersecurity service for the EU institutions, went further, advising any organisation that could not patch immediately to take its websites offline rather than risk being hacked. The European Commission, ENISA, CERT-EU, Europol, and the EU's network of national cyber response teams issued a joint statement urging immediate action, prioritising systems exposed to the internet.
What is PHP and why does it matter?
PHP, originally standing for Personal Home Page, was created in the early 1990s. It runs behind the scenes on roughly seven in ten websites, according to security firm Wiz Research, which calls it “a backbone of the web.” While HTML handles what you see on a page, PHP handles what you don't: checking logins, processing payments, serving up content. It underpins WordPress, Drupal, and Joomla, the platforms behind everything from personal blogs to online shops and government portals.
That reach turns two bugs into a global event. Wiz said every PHP version released in the past decade is affected by at least one of the flaws, and that even a bare-bones “Hello World” page on an unpatched server could be vulnerable. A fix was released on Wednesday, with the PHP Foundation telling tech departments worldwide that “if you run PHP, it is imperative that you update now.” Patched versions are PHP 8.5.12, 8.4.27, 8.3.36, and 8.2.35. Anything older no longer receives official updates and should be treated as exposed unless a vendor has backported the fix, CERT-EU warned.
The result has been a wave of apps and websites going dark, from the Paris police website to a Pokémon sleep tracker, as developers patch their underlying code. Government sites across Europe were among those pulled offline as a precaution.
How did AI find the flaws?
Both vulnerabilities, tracked as CVE-2026-109618 and CVE-2026-109619 and rated 9.2 out of 10 for severity, sit in the parts of PHP that nearly every request touches: memory management and error handling. If a web request hits PHP's memory limit at exactly the wrong moment, the software can be tricked into running an attacker's code without any login.
The PHP Foundation admits its own developers first spotted signs of the problem in 2024 during an external audit. A theoretical exploit was sketched out, but with no concrete route to an attack and a tiny core team weighing the trade-offs of a fix, work stalled and restarted for months. The foundation called the delay “a structural failure.”
Then AI got involved. In August, security firm Calif, working on an OpenAI-funded audit of PHP's source code and using OpenAI's GPT-5.6 Sol model, reported a fully working remote attack with the same root cause. OpenAI independently flagged the same issue the same day. Wiz followed with further attack vectors two weeks later, using its own Wiz Atlas tool alongside Google's Gemini 4 Argon and GPT-5.6 Sol.
“With frontier AI models and the right tools, a single researcher completed in a week what previously would have taken a world-class team months of work,” Wiz researchers wrote.
OpenAI called the discovery “one of the most impactful vulnerabilities in recent memory,” adding that it “marks a new phase, where AI can find security bugs with internet-scale impact.” The findings were kept under wraps while a fix was developed. Then, on Tuesday afternoon, the two CVE identifiers leaked onto online forums and social media. PHP brought its release forward by a day, and Calif mistakenly published—then deleted—a video showing its proof of concept working against several targets, the PHP Foundation said. Copies spread quickly.
“The trouble with tools like AI is that the moment all the information's out there, people can weaponise it very, very quickly,” Steve Kerrison, senior lecturer in cybersecurity at the Singapore campus of James Cook University, told AFP. “That's why everybody's acting with such urgency to respond to this issue.”
Have sites already been hacked?
Not necessarily. “Taking websites offline does not necessarily mean they have been hacked. It can be a responsible measure to protect systems while security updates are applied,” said Dimitrios Salampasis of Australia's Swinburne University of Technology. The PHP Foundation said its security team inspected the proof-of-concept code circulating after the leak and found no functional exploit at the time of release. “The evidence that security researchers are seeing is that nobody was actively performing attacks using the techniques available,” Kerrison said.
The picture is shifting, however. Web infrastructure firm Cloudflare said on Wednesday it had observed attack traffic aimed at exploiting the flaws. As European institutions and businesses race to patch, the incident underscores the double-edged nature of AI in cybersecurity: the same tools that can uncover hidden vulnerabilities can also be used to exploit them. For now, the advice from CERT-EU is clear: update your PHP versions immediately, and if you can't, take your site offline. The broader lesson is that AI is changing the speed of both discovery and attack, and Europe's digital infrastructure must adapt accordingly.
