Skip to content
Contact

Software Maintenance in the age of AI: Stable no longer buys you time

PHP2Shell showed how quickly a disclosed flaw can become a live attack. The gap between the version you run and a supported one is now, roughly, how long you have to patch.
A developer working at a desk with multiple monitors displaying code, in a modern shared office workspace.

How fast the clock now runs

PHP isn’t glamorous, but it sits underneath a remarkable amount of the internet. It’s often quoted as powering 70 percent of websites, and every PHP version from 5.4 onward was affected by at least one of these flaws. Between them, they could let an attacker run their own code on a server without ever logging in, under the right conditions. The Register

The fixes shipped at 14:00 UTC on 21 October, which was 3am on Thursday here in Aotearoa. By then, attackers already had a head start. AWS’s honeypots had logged an exploitation attempt four days earlier. Cloudflare’s CEO described thousands of attempts in the 10 hours before release, climbing to tens of thousands within 45 minutes of it, hundreds of thousands within two hours, and millions not long after. AWS Cloudflare

Of course, part of what made it so fast was AI. The researchers who found the bugs had used AI models to do it, and a security vendor who was told only that something was coming found the same flaws on its own overnight. AWS’s view is that exploitation used to wait for a small number of capable people to publish a working proof of concept, and that this dependency is weakening. The thing that’s so important to understand is that AI didn’t put these bugs in PHP. They’d been sitting there for years. What it changed is how quickly someone can get from knowing a bug exists to using it. The Register The Hacker News

The week it all got loud

If the speed was new, so was the noise. Much of it happened while New Zealand was asleep, so a lot of people here woke up to find it already in full swing.

“Coordinated disclosure” is meant to be a quiet process. The PHP team had spent weeks working with the researchers and with the people who package PHP for everyone else, so that the fix could land everywhere at once. That plan was held until two days before the planned release. Then the CVE numbers began circulating online with the wrong release date and a wrong severity score, and the PHP team brought the release forward by a day. Early the next morning, a distributor published fixed builds more than six hours ahead of the agreed release. The PHP Foundation and that vendor ended up arguing about it in public. Meanwhile, plenty of confident-sounding mitigation advice was doing the rounds online. Almost all of what we saw turned out to be well off the mark. The PHP Foundation and The Register. This was a genuinely extraordinary set of events and helps to explain the level of noise and public awareness which attached to this particular event.

For anyone responsible for a PHP system, all of this turned into a day of decisions made on rumour and conflicting advice, often before the facts had settled. Noise like that is part of the risk. When you can’t yet tell what’s true, the most comfortable place to be sitting … is one routine update away from safe.

What we did

You can read more about how this felt from the inside here.

Our own week started on Wednesday morning, when PHP’s advance notice landed. By Wednesday night we were discussing making the call to put our highest-impact sites behind a maintenance page before the patches had even been released. As a software consultancy, we look after other people’s products. Taking a site down when the owner can’t be reached is an extraordinary call, and we knew it wouldn’t be universally popular or understood. Taking a working site down when nothing has happened yet is a real cost, and it’s a hard conversation to have with someone whose service has just disappeared for reasons that still sound hypothetical. Within the first 24 hours, though, we saw an attempt to exploit the vulnerability against a site we manage, which made this real in a way it hasn’t been before. One of the takeaways for us is to speak with each of our partners now and agree a plan for permission in extraordinary circumstances like this in the future.

When the release came out at 3am, we had a team member up to read it and weigh the risk. By mid-morning we had patched deployments queued across our non-production environments, prioritised (a little grimly) perhaps, by thinking about which sites would make the worst headline if they were to be compromised.

The sites on older versions were the hard part, which is ultimately the point of this piece. PHP’s backport for 8.1 existed, but it didn’t install itself. Building and testing it was one job. A clash that stopped the patched build from running was another, and one of our engineers cracked it with an AI assistant alongside him. Happily, it was running in production the same day. Across the board, though, our nearly-but-not-quite end-of-life sites took longer and carried more risk to fix than everything else, and that cost traces straight back to decisions made over the preceding years to leave them be.

Happily for us, there were no outages or alerts overnight after the mass upgrade. After the week we’d had, that felt like its own small victory.

The part that surprised us

What we ultimately didn’t expect was how far the story would travel. Security advisories usually stay in technical circles, but this one made mainstream news. In the days that followed, partners got in touch with us, unprompted, to ask whether they were exposed. That has very rarely happened to us with a CVE before.

For all of them, the answer was no. Anyone who was affected had heard from us over Wednesday/Thursday. But the question still made us stop and think, because it had genuinely never occurred to us to tell anyone about a negative result. We’ve always been quick to tell clients when something affects them. But given just how many CVEs are discovered in a normal year, we’d never thought “this one doesn’t touch you, and here’s why” was worth saying too. Perhaps it is, and for events this size we’re now asking whether it should be our default. Supporting our partners to be more security aware is part of the work.

If all the coverage does one good thing, we hope it’s this. The importance of knowing where your systems are at, and what the risks could look like, isn’t a once-a-year audit item any more. It’s part of what it means to keep running software in this new climate.

PHP did more than it had to

It would be easy to read all of this as a story about PHP getting it wrong, and it isn’t. The PHP team got the fix to distributions and support vendors ahead of time, so most users had a simple way to update on the day. Because the bugs were so severe, it also published minimal source patches for 7.4, 8.0 and 8.1 (versions it no longer supports). Today's PHP Security Release, and How We're Strengthening PHP Security

It was goodwill under pressure, though, and not a promise. It’s worth being realistic about its limits. The older-version patches are meant for people who build PHP themselves or don’t have a support vendor. As we found, applying one is real work. Anything on 7.3 or older got nothing from PHP at all. And PHP 8.2, one of the versions that did get an official release, reaches end of life on 31 December 2026. Anyone who has just patched onto it has until the end of the year before they’re back in the same position. We advise against planning for this kind of generosity happening again. Help Net Security

What this changes

Those upgrades you deferred were never free, but the consequences are changing. PHP2Shell makes that day easy to picture: a leak overnight, a patch you can’t apply because your version is too old, and a decision about whether to switch off a working system while you catch up.

The useful question isn’t whether your platform still works. It’s how far you are from a supported version, and who in your organisation gets to decide to switch something off while you close that gap. If either answer is uncomfortable, now is the time to find out, while it’s still a choice, and long before you make the headlines for the wrong reasons.

Frequently asked questions

Is my site safe if I patched PHP for PHP2Shell?

If you’re on 8.2.35, 8.3.36, 8.4.27 or 8.5.12, or a vendor package that names both CVEs, you’re covered for these two flaws. If you were exposed before patching, check for signs of compromise from 17 October onwards. If you’re on 8.2, plan your next upgrade now, because it reaches end of life on 31 December 2026.

Which PHP versions got the PHP2Shell fix?

The PHP project released official fixes for 8.2 to 8.5. It published minimal source patches for 7.4, 8.0 and 8.1, which you build yourself or get through a distribution or support vendor. Versions 7.3 and older got nothing from PHP.

How do I know how exposed my platform is?

Start with what you run and how far each system is from a supported version. Then work out how long closing that gap would take if you had to do it in a week. A short, time-bound technical review can answer both questions before the next incident forces them.

Related

More Development Practices

Most Recent

Show all articles
Show all articles