AI helped find two serious flaws in PHP, a language behind many websites

What happened

PHP is a programming language that runs a lot of the web behind the scenes. When you log in to a site, search a shop or read a blog built on WordPress, there is a good chance PHP is doing the work on the server.

On 21 October, the PHP project published an emergency security advisory. It describes two flaws that could let an attacker run their own commands on a server, without needing a password. The advisory rates both at 9.2 out of 10, and says all PHP installations that serve web requests are at risk. It names the highest risk as servers using Apache with a module called mod_php, where the attack can work before any website code runs.

The flaws are serious, but the advisory does not say they have been used in attacks. It does say the team could release the fix earlier if the details became public or attacks began.

Who found the flaws

This is where accounts differ, so it helps to separate what is confirmed from what is reported.

Confirmed by PHP. The advisory gives the two identifiers and the fixed versions. It does not name anyone who found the bugs.

Stated by Wiz Research. Wiz, a cloud security company, published its own account on the same day. It says its researchers found the first flaw on 19 August, reported both to the PHP security team on 24 August, and worked with the PHP team until the fix was ready. Wiz says it used its own AI tool, Wiz Atlas, together with Google's Gemini 4 Argon and OpenAI's GPT-5.6 Sol. In its words, Gemini 4 Argon "made an invaluable contribution to finding the vulnerability". Wiz also thanks Calif and OpenAI for their work on discovery, coordination and checking the fix.

Reported by the press. Fortune and CNN report that Calif, a security firm working with OpenAI, found an earlier issue using GPT-5.6 Sol. We have not seen that claim in PHP's or Calif's own published account, so treat it as a report, not a confirmed fact.

So the honest answer to "which agent found it?" is this: Wiz says three AI tools helped, and more than one team was involved. No one has yet published a full, independent account of how each bug was found.

Why it matters

Most people will never see PHP. They will see the result if a site they use goes down, or if their password is exposed. The advisory says an attacker could read the memory of the PHP process, which can contain passwords, tokens and keys to other services. Wiz reports demonstrating the attack against the latest container images of WordPress, Drupal and Joomla, though it has not published the full details.

The story is also about AI and security. The same tools that help a defender find a bug can help an attacker find one. Wiz says frontier AI models "substantially reduced that barrier" between finding a flaw and using it. That is why the update matters so much: the faster a flaw is found, the less time there is to fix it before others find it too.

The idea behind it

If you are new to security, one idea helps here. A flaw is a mistake in the code, like a lock that opens for the wrong key. Finding one means looking for places where the code trusts input it should check. AI tools are good at reading large amounts of code quickly and spotting patterns, but a finding still has to be checked. That is the same logic as our story on AI and checkable answers: a tool can propose, but a test or a human has to confirm.

The loop is worth knowing too. An AI agent tries something, reads the result, and tries again, which is how it can search a large codebase on its own. What an agent is explains that loop in plain terms. And AI tools can still be confidently wrong, so a claimed flaw needs a reproducible test before anyone trusts it. Why models state falsehoods with confidence covers why.

What to do

If you run a PHP site, update now. The fixed versions are PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35. PHP 8.1 and older no longer get updates from the project. The advisory links minimal source patches for PHP 7.4, 8.0 and 8.1, but a supported newer version is the complete fix. The advisory also lists temporary settings that reduce risk until you can update, but it says they do not make a server safe. If you use shared hosting, ask your provider whether they have applied the update.

If you only use websites, you can help by keeping your own accounts strong and turning on two-step login where it is offered.

What to watch

• The full technical write-up. Wiz says it will publish more, and the PHP project's own GitHub advisories may name the finders.
• Whether attacks start. The advisory says the team could release details earlier if attacks appear.
• The 70% claim. Headlines say PHP runs 70% of websites. Wiz uses that figure without a cited source, so treat it as an estimate.