AI-Found PHP Flaws Put Websites at Risk Worldwide

Researchers using artificial-intelligence tools uncovered critical vulnerabilities in PHP, the open-source programming language behind an estimated 70% of websites where its use is known. Called PHP2Shell, the flaws could let attackers run code remotely, read sensitive memory, take control of servers, deface sites and steal customer data. That's a lot of exposure for people who simply shop, bank or file paperwork online.

The estimate doesn't mean every website is vulnerable. Recent data cited by the New York Times suggests PHP adoption is already waning. Still, the people most likely to feel the fallout aren't the engineers who wrote the code. They're small businesses, public agencies and everyday users whose information sits on servers they’ll never see.

How the Flaws Were Found

Teams at OpenAI, the Google-owned cybersecurity company Wiz and the cybersecurity company Calif worked on the discovery. Wiz researchers used Google's Gemini 4 Argon, Wiz Atlas and OpenAI's GPT-5.6 Sol. They found the bugs in late August and disclosed them publicly on Wednesday, October 21, after PHP had a fix ready.

OpenAI said it led a coalition of researchers, security providers and others who worked on the vulnerabilities over the previous two months. "PHP's maintainers led the fixes and release, with our teams and partners like Wiz, Cloudflare, and others contributing research, technical support, and coordination to prepare protections ahead of disclosure," the company said. The collaboration matters. Shared, open work on shared infrastructure is how the internet's commons get defended.

Who Bears the Risk

The PHP Foundation released patches and urged website operators to update immediately. But public fixes also hand hackers a roadmap. Experts said attackers could reverse-engineer patches and strike systems before operators protect them. That's the uncomfortable gap: people with the fewest resources to patch fast are often the ones most exposed.

Some government websites went dark. On Wednesday, the Finnish government's website was only partly available, with a message explaining that government sites were fully or partially unavailable while officials resolved the issue. UPI reported that governmental bodies in France, Denmark, Finland and the Netherlands took public-facing sites down for emergency updates. The New York Times reported that some European government sites became unavailable after agencies advised organizations to go offline if they couldn't apply a patch.

The National Security Agency advised organizations using PHP to update or take affected systems offline if they couldn't patch immediately. The FBI's cyber division made a similar recommendation. Britain's National Cyber Security Center urged organizations to "take immediate action to mitigate a security issue affecting PHP."

Official Response and Warnings

Japan's prime minister, Sanae Takaichi, posted a direct appeal on social media: "We have an urgent request for everyone to take action tonight. If it is difficult to apply the fix patch promptly, temporarily suspending websites and similar measures can be considered an effective way to reduce risk."

Cloudflare, which worked with the PHP Foundation, Calif, OpenAI and Wiz, said the flaws could permit remote code execution, disrupt services and expose customer data. Chief Technology Officer Dane Knecht was blunt. "We are seeing early attacks emerging and are updating mitigations as new vectors show up. This is a STOPGAP only. Patch as soon as PHP releases the update."

Ismael Valenzuela, vice president of labs, threat research and intelligence at Arctic Wolf, said the vulnerabilities put a "significant portion of the internet's business and commercial infrastructure at potential risk." He stressed that not all websites are vulnerable, but the flaws demand "immediate attention." Some weaknesses, he said, had existed for more than a decade. "What changed wasn't the underlying code, but the ability to find those weaknesses and turn them into working exploits."

The Bigger Question of Oversight

Wiz researchers compared the episode with the Heartbleed bug from 2014, which affected an encryption technology estimated to be used by two-thirds of the Web. They suggested PHP2Shell could prove a turning point. "With frontier AI models and the right tools, a single researcher completed in a week what previously would have taken a world-class team months of work."

OpenAI called the flaws "one of the most impactful security issues in recent memory," and warned it expects similar findings to rise sharply. The company also acknowledged that AI's power to find flaws could help attackers build exploits just as quickly.

Greg Brockman, president and second-in-command at OpenAI, wrote that the PHP fix process was slow and manual and wouldn't keep pace as more exploits emerge. He suggested governments might need a much larger role in monitoring internet infrastructure and possibly fixing vulnerabilities on other servers.

Raghu Nandakumara, vice president of industry strategy at Illumio, put the tension plainly. "We want AI to move quickly enough to find vulnerabilities humans might miss, while also making sure that increasingly autonomous systems operate within boundaries we understand and control." He added that speed without control creates a new kind of security problem.

Mother Jones reported that Anthropic had held back its Claude Mythos Preview after it became clear in April that the model could autonomously discover and exploit critical software vulnerabilities. Anthropic and OpenAI have given trusted partners, including major companies and government agencies, advance access to strengthen cyber defenses. Open-weight models, whose safeguards can be removed, have consistently lagged leading closed-weight models by about six to 12 months, the magazine reported, and hackers could use them against organizations with weaker defenses.

Why This Matters: The people who pay when a server goes down or a database leaks are rarely the companies that built the tools. They're patients whose records sit in a regional clinic's system, shopkeepers whose payment pages stop working, and citizens who can't reach a government service during an emergency update. Those costs hit hardest at organizations without the staff and budgets of large technology firms. That's a structural inequality in digital security.

Public institutions can't ignore the question this episode raises: when private labs and a few companies control the most powerful vulnerability-finding tools, who decides how fast the rest of the internet gets protected? Democratic oversight, shared public monitoring of critical infrastructure and clear accountability for how AI tools are released would give the millions who depend on these systems a real voice in that decision. Right now, many are learning about the risk the same way they learn about most cyber threats: when their sites go dark.