SDSignal Desk

AI finds two critical PHP flaws hidden since 2004, and the patch race begins

· Updated · Signal Desk newsroom

Analysis of reporting by TechCrunch. How we work

Image: TechCrunch

Calif, using OpenAI’s GPT-5.6 Sol, and Google-owned Wiz found critical PHP vulnerabilities present since PHP 5. The NSA told owners who cannot patch to take sites offline. Disclosure was responsible and no exploitation was reported, but Wiz warns AI now makes old bugs easier to weaponize.

Key takeaways

  • Two critical PHP flaws, present in every version since PHP 5, can let attackers reach a server’s file system or a connected database.
  • The NSA advised taking sites offline if they cannot be patched right away; the recommended target is PHP 8.5.12.
  • Calif found one flaw with GPT-5.6 Sol through OpenAI’s Patch the Planet program; Wiz found several with its Wiz Atlas agent, Gemini 4 Argon and GPT-5.6 Sol.
  • OpenAI says the bugs were found before attackers; there were no reports of exploitation in the wild.
  • Wiz warns large language models can now adapt exploits across PHP versions, configurations and memory layouts.

What was found

PHP sits under roughly 70% of websites, by TechCrunch’s count, including the servers behind WordPress. On Wednesday the NSA warned that two critical vulnerabilities, present in every version since PHP 5, let attackers reach a server’s file system or a connected database. Its advice was blunt: if a site cannot be patched right away, take it offline until it can. The PHP Foundation has released patches, and the recommended target is PHP 8.5.12. Anyone on PHP 8.1 or older is running unsupported software and faces real upgrade work rather than a quick fix. WordPress said the flaw is not specific to its software, but sites built on it still need their servers patched and should work with their hosts.

Security firm Calif, working inside OpenAI’s Patch the Planet program, used GPT-5.6 Sol to find one flaw. Wiz, now part of Google, found multiple vulnerabilities with its Wiz Atlas agent, Gemini 4 Argon and GPT-5.6 Sol. Both followed responsible disclosure, worked with PHP on fixes and brought in Automattic, Akamai, AWS, Cloudflare, Fastly, Imperva, Microsoft and Vercel to test. OpenAI says the bugs were found before attackers found them, and there were no reports of exploitation in the wild.

Discovery got cheap; exploitation may too

The defensive outcome is real. But according to Wiz, the bugs sat for two decades partly because they were too hard to weaponize, and its warning is that large language models can now adapt an exploit across PHP versions, configurations and memory layouts. Human effort was the barrier that kept old bugs harmless, and that barrier is shrinking: the capability that surfaced the flaws also lowers the cost of using them. OpenAI has said its Astra model was the first to reach the critical cybersecurity threshold in its Preparedness Framework and found zero-days during evaluation (Path to Astra), and Google is pitching Gemini 4 Argon on cyber defense (Gemini 4 Argon).

The episode lands the same week OpenAI published a cyber-defense policy blueprint arguing that defenders are in a temporary window of advantage.

The long tail of patching

The bottleneck has moved from finding bugs to fixing them across tens of millions of servers. Small businesses on shared hosting, abandoned sites and appliances nobody remembers running PHP make up a long tail where damage is likely to land, and the burden of pushing updates falls largely on hosting providers. The NSA’s advice to go offline rather than wait signals how short the patch window may be now that details are public. Security teams should prioritize internet-facing PHP applications with database access and watch for exploit attempts.

AI-driven discovery under responsible disclosure is preferable to less careful actors finding these flaws first, but it only works if patch distribution scales as quickly as discovery. That may require hosts to push updates by default and policymakers to treat end-of-life runtimes as a security liability. More in our OpenAI coverage, Google coverage and cybersecurity coverage.

Who feels it

Site owners and small businesses
Check PHP versions with hosts now; 8.1 or older needs an upgrade path.
Hosting providers
Most of the burden of rolling out PHP 8.5.12.
Security teams
Expect exploit attempts on unpatched internet-facing servers.
AI labs
A showcase for responsible discovery, and a shorter disclosure-to-exploit window.

What to watch

  1. First confirmed in-the-wild exploitation of the PHP flaws.
  2. Patch adoption figures from major hosts and CDNs.
  3. More legacy runtime bugs surfacing through AI programs.
  4. Federal guidance on end-of-life software as a liability.

FAQ

What PHP versions are affected?

Every version since PHP 5; the recommended patched target is PHP 8.5.12.

Who found the PHP vulnerabilities?

Calif, using OpenAI’s GPT-5.6 Sol, and Google-owned Wiz, using its Wiz Atlas agent, Gemini 4 Argon and GPT-5.6 Sol.

Have the flaws been exploited?

There were no reports of exploitation in the wild at disclosure.

What did the NSA advise?

Patch immediately, or take sites offline until they can be patched.

Read the original

Continue at the source.

TechCrunch

Companies: OpenAI, Google

More on OpenAI

All OpenAI coverage →