Skip to content
Cyberttak.
· 3 min read

Two PHP flaws hidden for decades were turned into working attacks with AI help, and millions of exploit attempts followed

Wiz says nearly every PHP version since 5.4 is affected, and patches arrived on October 21, 2026, as governments took some sites offline.

By CyberttakPublished · 16:32

In brief

AI helped crack PHP flaws hidden for decades

Two critical flaws affect nearly every PHP version since 5.4. Patches came October 21. Cloudflare saw millions of exploit attempts, so website owners should update now.

A network technician seen from behind in a doorway plugs a grey cable into a small switch on a bare wall beside a blue crate
A network technician seen from behind in a doorway plugs a grey cable into a small switch on a bare wall beside a blue crateCredit: AI-generated · FLUX.2 klein 9B

Frontier AI models helped security researchers turn two old, hidden flaws in PHP into working exploits. That is the account from Wiz, a security firm owned by Alphabet, in its own write-up. The flaws are tracked as CVE-2026-109618 and CVE-2026-109619, and nicknamed php2shell.

PHP is the programming language behind a huge share of websites. Wiz says it powers 70% of them, including WordPress, Drupal and MediaWiki, the software behind Wikipedia.

Wiz says frontier AI models substantially lowered the barrier to turning a longstanding flaw into a working exploit.

What the flaws allow

Both flaws involve reaching memory without permission. An attacker can then run commands on a server over the internet, with no password and no account. Wiz also says the bugs can leak memory holding passwords, tokens, API keys and certificates.

Wiz rates severity as high as CVSS 9.2. It says nearly every PHP version, going back to 5.4, is affected by at least one of the flaws.

Where AI came in

Wiz says the flaws had sat in the code for decades. The hard part was proving they could be exploited. That took a chain of steps Wiz describes as at the edge of what human researchers could practically do. Frontier AI models lowered that barrier, according to Wiz.

Fortune credits OpenAI's GPT-5.6 Sol model with a key role.

The timeline, and where sources differ

Al Jazeera reports that Calif, working with OpenAI, reported the first flaw on August 11, and Wiz reported the second on August 24. Wiz's own timeline gives August 24 as its report date and September 21 as when PHP began a fix.

Patched PHP releases came out on October 21, 2026, according to Wiz. Fortune says the flaws became public on October 20, after reports broke late that day, ahead of the official release.

The rush of attacks

Per Fortune, Cloudflare counted tens of thousands of exploit attempts in the first 45 minutes after word spread. Within two hours the figure was in the hundreds of thousands, and by the evening of October 21 it had grown to millions.

Amazon Web Services said it saw attempts days before the October 20 disclosure, which suggests word may have leaked early. Fortune reports that claim.

Governments reacted fast. Al Jazeera says Germany, France, Belgium, the Netherlands, Finland and the Czech Republic took some sites offline. The US National Security Agency advised taking sites down if they could not be updated at once. The PHP Foundation urged every site admin to update immediately.

If you run or use a website, the practical point is simple: the patch matters more than the headline. Is your site's host on the new version?

Limits

The sources show attacks on these flaws. They do not confirm that AI drove the attackers. CNN describes AI helping researchers build exploits. Wiz's blog is the only first-party source here; no PHP Foundation or OpenAI page was available.

Sources