Money and business in the Middle East.

AI

AI uncovers worst PHP flaw in a decade, hitting 70% of websites

A critical vulnerability that could let hackers take over any system running PHP was disclosed Wednesday by the technology’s creators, who warned it is the most severe issue the project has faced in more than a decade.

· Originally published by ontime+ · Last verified: 21 Oct 2026 (Sukaina Khalid)

Key Points

  1. PHP disclosed PHP2Shell, a critical memory flaw letting attackers seize control of unpatched servers.
  2. Researchers at security firm Calif found it using OpenAI's GPT-5.6 Sol during a funded code audit.
  3. PHP runs as much as 70% of the world's websites, making patching speed the decisive variable.

The latest:

A critical vulnerability that could let hackers take over any system running PHP was disclosed Wednesday by the technology’s creators, who warned it is the most severe issue the project has faced in more than a decade. Dubbed PHP2Shell, the flaw sits in how PHP handles computer memory and can be abused to run unauthorized software on unpatched web servers. PHP powers as much as 70% of the world’s websites.

Details:

  • The flaw: PHP2Shell is described as a class of problems in PHP’s memory management rather than a single bug, exploitable in multiple ways. A successful attack would let an intruder steal information or launch a cyberattack against visitors to a compromised website, according to the disclosure.
  • Speed of attack: Researcher Ryan Austin found he could control code running in a server’s memory with as few as a few dozen requests to a machine running PHP. On smaller systems the hack takes seconds; on more complex ones it can take hours, according to Calif.
  • The AI role: Austin uncovered the flaws with help from GPT-5.6 Sol, one of OpenAI’s most advanced models, during an August audit of PHP source code. OpenAI funded that audit and others as part of an effort by AI companies to help open-source projects address vulnerabilities now surfaced by AI systems.
  • Hidden for years: The class of bugs had sat in PHP’s source code for years, said Calif chief executive Thai Duong. A researcher flagged the area decades earlier, but the flaw was judged too difficult to exploit and was largely forgotten until AI found a path.
  • The comparison: Volker Dusch, a security engineer with the nonprofit foundation managing the PHP project, called it the most severe issue in over a decade and likened it to Heartbleed, which forced a global race to protect hundreds of thousands of servers more than ten years ago.
  • Wiz assessment: Ami Luttwak, co-founder of Google’s Wiz cybersecurity team, called PHP2Shell the most severe AI vulnerability ever found, and said it was likely too complex to have been discovered without AI. Wiz researchers independently developed new exploits and shared them with the group, alongside findings from OpenAI’s research team.
  • The patch scale: Close to 200 files were fixed and developers wrote more than 10,000 lines of code during remediation, carried out by an informal working group of security researchers and engineers coordinating with software distributors and security companies.
  • The leak: A patch date had been set for October 22, with the flaw kept a closely guarded secret. News of it leaked Tuesday in a post to Hacker News, just under two days before release, pushing Dusch’s team to rush out the update and warn administrators to expect a critical fix Wednesday.
  • Administrator response: Web administrators are scrambling to protect their systems, and some say they are pre-emptively taking their sites offline. Duong warned that someone could run AI against the published patches and post an exploit script, saying “all hell will break loose.”

Background:

Patch releases traditionally start a race: defenders test and install fixes while attackers study the changed code to reverse-engineer an attack. Security experts say today’s AI tools compress that attacker timeline far below human speed.

Between the lines:

The same capability sits on both sides of this story. AI surfaced a flaw human reviewers had dismissed as unexploitable for decades, and the patch across roughly 200 files is now a map attackers can feed to the same class of models. The rushed release after the Hacker News leak reflects that asymmetry: Dusch framed it as informing defenders without handing attackers extra time.

What’s next

Watch whether a working public exploit script appears in the days after Wednesday’s patch, how fast hosting providers and distributors push the update downstream, and whether Wiz or OpenAI disclose additional PHP2Shell exploit paths.

Read on ontime+ ↗

AI uncovers worst PHP flaw in a decade, hitting 70% of websites · INXEN