More to Come: AI Found PHP2Shell, and OpenAI Says There Are More. Is Your PHP Ready?

What OpenAI said

A few hours after PHP released its fixes on 21 October 2026, OpenAI announced that its models had helped find them:

Due to the severity of these flaws and PHP’s widespread use, we consider this bug to be one of the most impactful vulnerabilities in recent memory. It also marks a new phase, where AI can find security bugs with internet-scale impact.

Then came the line that got everyone’s attention, picked up by AI commentator Andrew Curran as “the tip of the iceberg”:

As significant as they are, these bugs are among several significant vulnerabilities we’ve already found and are working to fix, and we expect similar findings to increase sharply in the coming months.

OpenAI hasn’t said what the other vulnerabilities are, or which software they affect. But the message for anyone running PHP is clear: PHP2Shell won’t be the last emergency of this kind, and the next one may not be far away.

How PHP2Shell was found

According to OpenAI, researchers at the security firm Calif, working under its Patch the Planet initiative, used the GPT-5.6 Sol model to find a version of the bug in August. OpenAI then spent two months leading a coalition of PHP maintainers, researchers and security companies to prepare the fix, which it says was “quite intricate and required a major change to the PHP engine”. Wiz Research reported the flaws too, and PHP’s security advisory credits several researchers.

The plan was a coordinated release on 22 October. Details leaked early, so PHP’s maintainers brought it forward to 21 October, announcing it the evening before.

OpenAI calls this the “Defender’s Window”: AI found the flaw before attackers did, which gave defenders time to prepare. For two months, the people who maintain PHP, Linux distributions and hosting platforms were getting ready. The rest of us got about a day.

This isn’t a one-off

PHP2Shell is the second time this year that AI has found a critical flaw affecting a large share of the PHP web:

With wp2shell, Wiz saw multiple attackers successfully planting webshells on unpatched WordPress sites within days of disclosure. With PHP2Shell, The Record reported that watchTowr was already seeing exploitation attempts on release day, using proof-of-concept tools built with AI help, and Cloudflare said it was seeing attacks too. Government, hospital and company websites across Europe were taken offline as a precaution.

AI is speeding up both sides. Defenders find bugs sooner, and attackers turn a published fix into a working exploit faster than before.

The catch: the head start only helps if you can take the fix

The Defender’s Window works like this: a flaw is found, a fix is prepared quietly, and on release day everyone updates. That last step assumes your site is running a PHP version that gets the fix.

For PHP2Shell, PHP 8.2, 8.3, 8.4 and 8.5 got patched releases. PHP 8.1 and older got none. According to Detectify, quoted by The Record, roughly half of PHP websites run versions that are no longer patched. Those sites got no benefit from the two months of preparation. Their owners were left with configuration workarounds, unofficial patches or switching the site off.

Detectify also warned that attackers are building lists of PHP servers sorted by version. An outdated PHP version is easy to spot from outside and slow to fix from inside, so it’s exactly what attackers look for.

If OpenAI is right that these findings will “increase sharply”, this stops being a rare bad week. It becomes a regular event, and each one works the same way: supported versions get a fix within a day, and everything else is exposed.

How to be ready for the next one

You can’t predict what will be found next. You can make sure that when it is, patching takes an afternoon rather than a crisis meeting.

1. Run a supported PHP version, with time to spare

PHP 8.2 reaches end of life on 31 December 2026. If you’re upgrading, go straight to PHP 8.5, which is supported until the end of 2029. On PHP 7.4, 8.0 or 8.1, this is now the most important item on your list.

2. Know where PHP runs

Make a simple list: every server, container, hosting account and appliance running PHP, its version, and who is responsible for updating it. The NCSC’s first instruction for PHP2Shell was to find every PHP installation. Doing that in the middle of an incident is slow.

3. Make patching routine

If installing a PHP update means a nervous manual change on a live server, it will be put off. Automated tests, a staging copy and a repeatable deployment turn “update PHP and restart” into a ten-minute job you can do the same day.

4. Keep dependencies current too

AI is also scanning frameworks, plugins and Composer packages. wp2shell was in WordPress itself. Run composer audit and composer outdated regularly, keep plugins updated, and remove anything abandoned. Packages also drop support for old PHP versions, so an outdated PHP version eventually stops you getting their security fixes too.

5. Shrink what’s exposed

Turn off PHP features you don’t use: PHP2Shell’s second flaw went through session upload progress, which is on by default and rarely needed. Limit request and upload sizes at the web server, and don’t accept file uploads where the site doesn’t need them.

6. Hear about it early

Watch the PHP security advisories, and if you’re a UK organisation, sign up for the NCSC’s free Early Warning service. A WAF such as Cloudflare’s can block known attack patterns while you patch. It doesn’t replace patching.

The bottom line

For years, running an old PHP version was mainly a slow-burning risk that rarely seemed urgent. AI-assisted vulnerability research has changed that. Serious flaws in widely used software are being found faster, published in bigger batches and turned into exploits sooner. The sites that stay safe will be the ones that can take a fix on the day it ships.

Upgrading PHP is rarely the rewrite people expect. Most of the work is finding deprecated code and outdated packages, and tools like Rector and PHPStan handle much of it. It’s far easier to plan now than during the next alert.

Frequently asked questions

Was PHP2Shell really found by AI?

According to OpenAI, researchers at the security firm Calif used its GPT-5.6 Sol model to find the flaws as part of its Patch the Planet initiative, about two months before the fix was released. Wiz Research also reported them, and PHP’s advisories credit several reporters. The fix itself was written and reviewed by PHP’s own maintainers.

What other vulnerabilities has OpenAI found?

OpenAI hasn’t said yet. It describes the PHP bugs as “among several significant vulnerabilities” it has found and is working to fix, and expects similar findings “to increase sharply in the coming months”. Earlier in 2026, the wp2shell flaws in WordPress core were also found with the same model, by Searchlight Cyber.

Doesn’t AI help defenders more than attackers?

It can, if defenders can act on what it finds. Finding a flaw early only protects you if a fix is released for your version and you can install it quickly. Sites on end-of-life PHP versions get no official fix, so the head start goes to waste. Attackers also use AI: on the day PHP2Shell was published, watchTowr reported exploitation attempts using proof-of-concept tools built with AI help.

Which PHP version should I be on?

PHP 8.5 for anything you’re upgrading now. It gets security fixes until 31 December 2029. PHP 8.3 and 8.4 are still supported, and PHP 8.2 is supported only until 31 December 2026. Anything older than 8.2 no longer gets official security fixes.

Is a WAF or Cloudflare enough protection?

No. A web application firewall can block known attack patterns and buys useful time, but rules are written after a flaw is known, they can be bypassed, and some flaws can’t be blocked at that layer at all. Treat a WAF as a stopgap while you patch, not instead of patching.

Based on OpenAI’s announcement, PHP’s security advisories and published reporting as of 22 October 2026. We’ll update this article as more findings are announced.