In a dark room, people sit with their backs to the camera at desks with desk lamps, phones and mugs, facing computer monitors and a wall of large screens glowing with blurred patches of blue, amber and red.

Security

AI Helped Find Two Critical Holes in PHP

PHP fixed two flaws, each rated 9.2 out of 10, that could let attackers run code on web servers; OpenAI says its model helped find one.

SAN FRANCISCO, Oct. 21 — The PHP project on Wednesday released fixes for two critical flaws in its engine that can let an attacker who has not logged in run code on a web server. “All PHP users are strongly encouraged to update immediately,” its advisory said.

OpenAI said one of the flaws was found with its GPT-5.6 Sol model, and that it expects similar findings “to increase sharply in the coming months.” Calif, a security firm working in OpenAI’s Patch the Planet initiative, made that find, OpenAI said, and Wiz, a cloud-security company, independently found both.

The flaws sit in PHP itself, so even a simple “Hello World” application on an affected version can be vulnerable, Wiz wrote. Wiz and OpenAI say PHP runs about 70 percent of websites; W3Techs, which surveys the technology websites use, counts it on 69.3 percent of those whose server-side language it knows.The Old VersionsBy W3Techs’s count, 34.7 percent of websites that use PHP run version 7 or older. The PHP project no longer updates those versions, though for these flaws it published minimal source fixes for 7.4. The risk is highest, PHP said, on Apache with mod_php. Both flaws are rated 9.2 on CVSS 4.0, a severity scale that runs to 10.

The flaws’ CVE numbers leaked online on Tuesday, so PHP moved its release up a day, the PHP Foundation said. After the leak, it said, Calif mistakenly published a video of its proof of concept working against a number of targets; it was deleted, but copies spread and helped speed the development of various proof-of-concept exploits. Calif’s write-up does not mention the video. At release, PHP’s security team had found no functional exploit among those it inspected after the leak, the foundation said.

OpenAI wrote on X that the bug “marks a new phase, where AI can find security bugs with internet-scale impact.” In its blog post, it warned that the same capabilities “can also help attackers rapidly develop offensive tools and exploits.” It said it had engaged the White House’s Super Intelligence Force, the Office of the National Cyber Director and CISA, and pointed to $1 billion of subsidized access to its Daybreak cyber models, pledged in September.

OpenAI’s account is one of several. Wiz said it used its own Wiz Atlas alongside Google’s Gemini 4 Argon and GPT-5.6 Sol, and that Argon “made an invaluable contribution to finding the vulnerability.” Calif’s write-up names no model and says Calif reported “the second bug” to PHP too, on Aug. 21. Its researcher, Ryan Austin, wrote that the first find used no special harness and no model that wasn’t generally available, and that he knew the target’s vulnerability history well. “No model is simply capable of unearthing ‘all bugs’ with simple prompts,” he wrote. “At least for now.”

The PHP Foundation, which pays developers to work on PHP’s security, said a developer it contracted had found the first signs of the problem behind one of the flaws in 2024, but that work on a fix stalled for months. “The delay was a structural failure,” it wrote. Calif and OpenAI each reported the same root cause on Aug. 11, it said.

“Tools now find bugs faster than ever, for defenders and attackers alike,” the foundation wrote. In a tally published Oct. 14, the PHP security team, which runs separately from the foundation, said it had published 15 CVEs since July 1 and closed 106 reports that were not vulnerabilities, duplicated others or did not reproduce. The foundation thanked OpenAI and Anthropic for tools and funding and asked those who depend on PHP to pay for more engineers.

The fixed versions are PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35. There are no complete mitigations, PHP’s release notes said: “updating is the only safe choice.”

Daily Cortex is written with AI models, including Anthropic’s Claude, and fact-checked with OpenAI’s GPT models.

Sources

  1. PHP, “Security advisory: Critical remote code execution vulnerabilities, update now,” Oct. 21, 2026.
  2. PHP, “PHP 8.5.12 Released!” Oct. 21, 2026.
  3. PHP (GitHub security advisory), “RCE via OOM corruption (CVE-2026-109618, GHSA-w8hw-w9fx-j79w),” Oct. 21, 2026.
  4. PHP (GitHub security advisory), “RCE via stale global state (CVE-2026-109619, GHSA-6wmf-4rv5-p3rp),” Oct. 21, 2026.
  5. The PHP Foundation, “Today’s PHP Security Release, and How We’re Strengthening PHP Security,” Oct. 21, 2026.
  6. The PHP Foundation, “How the PHP Security Team Handles Your Reports,” Oct. 14, 2026.
  7. OpenAI, “Patch the Planet: global response coordination as agents uncover critical vulnerabilities,” Oct. 21, 2026.
  8. OpenAI, on X, “Post on the PHP vulnerabilities,” Oct. 21, 2026.
  9. Wiz Research, “PHP2Shell: AI uncovers critical RCE vulnerabilities affecting 70% of the Internet,” Oct. 21, 2026.
  10. Calif, “A tale of two PHP remote vulnerabilities,” Oct. 21, 2026.
  11. OpenAI, “Daybreak for Frontline Defenders: $1B to protect essential services,” Sept. 3, 2026.
  12. W3Techs, “Usage Statistics and Market Share of PHP for Websites, October 2026,” Oct. 22, 2026.
  13. FIRST, “Common Vulnerability Scoring System Version 4.0: Specification Document,” Nov. 1, 2023.