The PHP incident offers some important lessons for IT leaders

In October, an update was issued for two critical issues in PHP, the programming language powering a large share of the world’s websites, after a working exploit was published online.

The result of attackers exploiting the flaws was deemed so serious that the UK National Cyber Security Center (NCSC) advised those who couldn’t patch to take their systems offline.

What followed was a chaotic scramble as firms tried to apply the patch as quickly as possible.

A few weeks on, as the dust starts to settle, some experts believe the threat was overblown. Why is that, and ultimately, what lessons can be learned?

Volume driver

The PHP flaws were discovered in August by researchers at cybersecurity company Calif. “They used an AI model to help, but a human decided where to look, building on a similar bug first found back in 2004,” explains Daniel Card, cybersecurity consultant at Xservus.

Tracked as CVE-2026-109618 and CVE 2026-109619, the flaws could allow an attacker to crash a website, read its memory or, in some setups, take control of the server. Both issues were rated critical, carrying CVSS score of 9.2 out of 10, due to their ability to allow attackers to execute code remotely (RCE) without needing to authenticate.

Yet the disclosure of PHP2Shell didn't go to plan. PHP planned a coordinated fix, but the details leaked two days early and one company released its patch ahead of time. This led to the fix being rushed out on 21 October.

Sign up today and you will receive a free copy of our Future Focus 2026 report - the leading resource for IT decision-maker insight on priorities and investment areas in AI, security and more.

“By the time the PHP Foundation released its advisory on 21 October, attackers already had enough information to begin reverse-engineering the fix, with scanning and exploitation attempts observed within hours,” explains Howard Hughes, managing director, cybersecurity at LRQA.

Practitioner experiences

Patching PHP was a major challenge for many of the practitioners trying to address the bugs.

Some organizations took their websites offline as a precaution, including the Metropolitan Police, Dutch government services, and parts of Wikipedia.

Even national cyber agencies went dark. Australia’s Australian Signals Directorate (ASD) took down cyber.gov.au, the very site carrying its guidance.

Patching was often more difficult than it sounds, says Card. “People on shared hosting had to wait for their provider, while older PHP versions got no official fix.”

However, some back porting occurred, where a security fix is rolled into older versions.

Card, who live-patched his WordPress site on X, says “not exposing management interfaces to the internet goes a long way to reducing the likelihood of compromise via that route”.

For many teams, the hardest part was confirming they were safe after patching. “Version numbers proved unreliable, and hosts with expose_php switched off never advertised a version at all,” says Christiaan Beek, VP of intelligence at Rapid7.

Meanwhile, teams relying on external scans “got misleading answers in both directions”, according to Beek.

Current state of affairs

It is now around three weeks since the PHP patch. According to reports, there have been millions of attack attempts, but only a handful of confirmed break-ins. These have mainly centred around Palo Alto firewalls where admin interfaces were open to the internet, says Card. “I have seen no intelligence suggesting mass exploitation – although nobody can see the whole internet, and earlier break-ins may still come to light.”

Indeed, experts think the issues ended up being less impactful than they originally seemed, due the difficulty in actually performing an attack. Successfully exploiting the vulnerabilities in a live environment is more challenging than demonstrating an attack in a lab, points out Hughes. “A working attack requires information about the memory layout of the target's specific PHP build, which cannot easily be determined remotely without crashing the process.”

Yet while patching addresses the vulnerability, it doesn't undo what may have happened before the update was applied, says Hughes.

In the case of PHP2Shell, attackers could potentially access sensitive information, including session tokens, passwords, API keys, and database credentials, without needing to achieve full remote code execution, he says. “This means servers exposed before patching should be treated as potentially compromised.”

On a positive note, the PHP findings did show how AI can help researchers work out how attackers might exploit software flaws before they are able to do so in real-life.

AI is helping experts find serious bugs more quickly, even in old software everyone assumed was safe, says Card. It’s also helping them build working exploits faster, at least in lab conditions, he says. “Turning a lab exploit into one that works reliably against real-world systems is still hard, because many servers are set up differently, and software is compiled in different ways.”

With PHP2Shell, the fixes came out quickly, Card points out. But, according to Card, the real problem was poor information, which left people unable to judge their actual risk.

In inexperienced hands, AI “mostly produces noise that drove so much of the panic”, says Beek. He points out that a large share of the ‘proof-of-concept’ exploits circulating were “clearly AI-generated by people without the subject matter expertise to make them work”.

The PHP patch does offer lessons to prevent future patching frenzies as AI finds flaws at scale.

Most vulnerability response plans assume advance notice before a patch day. This case gave “almost none”, says Hughes. He thinks this is “a good moment to test whether your plan works with zero notice”.

Organizations need to understand what they've found, what they've fixed, what they're still verifying, and what they're relying on suppliers for, says Hughes. “Ultimately, ‘we patched’ is not the same as ‘we're assured’.”

It’s also worth noting that the frenzy was driven by headline numbers that “measured the wrong thing”, says Beek. “Exposure counts and exploit attempts ran into the millions, yet neither tells you how many servers were compromised.”

Beek believes organizations should judge risk by evidence of execution, correlating suspicious requests and abnormal worker crashes with unexpected shell execution, outbound connections, and payload downloads from web-server processes.

At the same time, leaks and hype will happen again, so the answer is always “good foundational security”, rather than “faster panic”, says Card. “That means assuming unknown flaws already exist, using layers of defense, and locking servers down so a single bug gains an attacker very little. It also means automating patching, knowing what software you run, and watching for signs of a break-in. None of this is new advice, but most organizations don’t follow it.”

The challenge is that good foundational security – often called basic – is “not always basic or easy to put in place”, explains Card. “But that should not scare people off. It’s not impossible and the more effort you put in, the better results you get.”

Original source The PHP incident offers some important lessons for IT leaders

Back to home