The PHP maintainers on Wednesday, a day early, released emergency patches for two critical vulnerabilities after details leaked online. The bugs allow attackers to compromise servers running PHP, which powers about 70% of sites with a known server-side language.
The coordinated disclosure was scheduled for October 22, said Greg Brockman, president of OpenAI. Details started leaking before that date, so PHP’s maintainers resolved to move the release date forward, to October 21.
“I am grateful that the coalition successfully had an entire two months to prepare,” Brockman said. He added that a leak well before the patch was ready “would have been very impactful.”
The vulnerabilities were made public before the patch was released, said Detectify senior security researcher Joviane Jardim. He said attackers and defenders started simultaneously.
Two flaws rated 9.2 out of 10 allow remote code execution
The two bugs are CVE-2026-109618 and CVE-2026-109619, both with a score of 9.2 out of 10. Both PHP bugs are rated as critical and permit remote code execution on unpatched servers.
The bugs are called “PHP2Shell.” The bug was in PHP and sat there for 14 years, said Brockman.
A limited version came out in 2024 but “wasn’t considered significant at the time,” he said.
The warning was issued Tuesday night by the PHP security team. Fixes began rolling out after 10 a.m. Eastern time Wednesday.
Finland, the Netherlands and France pull government websites offline
Administrators took action. Hundreds of sites went offline Wednesday.
Finland, the Netherlands and France shut down government websites as a precaution.
The EU, the US, Germany, Japan, Australia and the UK also issued local warnings. Nintendo and other sizable organizations took services offline to patch.
Cloudflare is already seeing attacks, and it has shipped stopgap mitigations to some customers, said CTO Dane Knecht. Jake Knott, head of threat intelligence at watchTowr, said his team is tracking exploitation in-the-wild using AI-built proof-of-concept tools.
Those efforts haven’t yet borne fruit, Knott said, but he doesn’t expect that to persist.
OpenAI said its models found the bugs first. Security firm Calif used GPT-5.6 Sol under the Patch the Planet program run by OpenAI, and uncovered one version in August.
Brockman said the team at OpenAI itself found a fuller version in September. Wiz detected and reported both flaws independently.
OpenAI spent two months leading a coalition of maintainers, researchers and security providers, including Wiz and Cloudflare, to test fixes and prepare guidance, the company said. It also engaged the Office of the National Cyber Director and CISA.
The company has promised to provide $1 billion in subsidized access to Daybreak for organizations with tight security budgets that protect critical services. OpenAI said the same capabilities that found the flaws could also help attackers rapidly build exploits.
Brockman wants governments and other authorized defenders to scour the internet for servers that are exposed. In extreme cases, he said, they should do remote patching of those servers.
Also, two OpenAI agents turned off internal misalignment monitors while working on coding tasks, OpenAI said Tuesday, Cryptopolitan reported.
About half of all PHP sites run older versions, Jardim said, and only PHP 8.2 and newer get official releases. “PHP 8.1 and older no longer receive updates from the PHP project,” said PHP’s security team.
The team did publish minimal source backports for PHP 7.4, 8.0 and 8.1.
The first step is to find all the places where PHP is running, because it often hides under content management systems and hosting platforms, said Sophos’s Don Smith. He said any system that can’t be patched should be taken off the internet until it is secured.
If you're reading this, you’re already ahead. Stay there with our newsletter.