TL;DR · 30-second read
The Short Version
A hidden weakness in PHP, the programming language behind roughly seven in ten websites, could let attackers take over web servers without a password. Some government websites in the United Kingdom, Finland and Japan went offline while they were fixed.
The surprising part: the problem had been known since at least 2024, but the fix stalled because it looked too hard for attackers to use. Artificial intelligence tools helped show how it actually could be used.
The lesson: when machines can find the way in quickly, the people running websites have to fix problems just as quickly.
The Associated Press reported that the PHP Foundation and the FBI’s cyber division warned on Wednesday of two critical vulnerabilities in PHP, the scripting language that powers about 70% of websites globally. The flaws could give attackers unauthenticated access, meaning remote control of a web server without a password. Advisories urged users to patch immediately or shut down sites they could not update, and government services in the United Kingdom, Finland and Japan were disrupted as sites went offline.
The vulnerabilities had been known since at least 2024, but work on a fix stalled because no practical way to exploit them was apparent. Humans first discovered the flaws; AI agents later helped identify how they could be exploited and helped test the fixes, in what OpenAI described as a coalition effort over the past two months. Researchers at cloud security platform Wiz said they independently used AI to identify the vulnerabilities.
Executive Summary
Two critical flaws in PHP have triggered one of the broadest emergency patching efforts in recent memory. Because PHP sits underneath WordPress, Drupal and MediaWiki, the software behind Wikipedia, the exposure reaches government portals, corporate sites and hosting platforms alike. Some organizations chose downtime over risk, and some sites stayed offline while waiting for software-specific patches and testing them.
The more lasting significance is how the flaws became urgent. They were known for at least two years but deprioritized because exploitation looked impractical. AI-assisted analysis helped establish that they could in fact be exploited. For data center, hosting and cloud operators, that turns every known-but-parked vulnerability into a question worth reopening, and it compresses the time they have between disclosure and deployment of a fix.
The ‘Unexploitable’ Backlog Just Lost Its Grace Period
The most consequential detail here is the timeline, not the AI label. The PHP Foundation said the flaws had been known since at least 2024. Its security team developed several competing fixes, each with downsides, and because there seemed to be no practical, concrete way for attackers to exploit the bugs, progress stalled. That is a common and usually rational triage call. Security teams rank issues by how realistically they can be attacked, and a weakness with no known attack path waits behind ones that have one.
AI agents then helped identify how the vulnerabilities could be exploited, and helped test and validate the fixes, as part of what OpenAI described in a blog post as a coalition of researchers and security providers working on PHP over the past two months. The mechanism matters. The assumption that kept these bugs on the back burner — that nobody could practically use them — was the one machine-assisted analysis overturned. Once it fell, a two-year-old issue became an advisory urging operators of a language behind about 70% of websites to patch or go offline.
For data center, hosting and cloud operators, the implication is specific. Any internal backlog of known issues deprioritized because exploitation looked impractical now carries a risk rating that may be stale. If defenders’ AI tools can find an attack path, attackers’ tools can too, and the gap between “theoretical” and “urgent” can shrink to the length of a single disclosure cycle.
Patch or Shut Down: The Operator’s Real Choice
The advisories gave site owners two options: apply the patches, or take sites offline if they could not update immediately. Several organizations chose downtime, and government services in the United Kingdom, Finland and Japan were disrupted. For anyone running PHP at scale, that is the real cost of a compressed patch clock: every hour a fix is not deployed is either exposure or an outage.
Patching is not one action. There are multiple patches for different types of software, some organizations were still waiting to receive theirs, and each has to be tested before it goes live. Shared and managed hosting providers sit at the sharp end, because a single provider may run many customer sites on a mix of WordPress, Drupal, MediaWiki and custom code, each with its own update path. An unauthenticated remote-access flaw also matters beyond the individual site: James Sullivan, director of cyber and tech at the Royal United Services Institute, noted that a compromised server could be used as a platform to launch other attacks.
Charl van der Walt, head of security research at Orange Cyberdefense, called the reaction so far the “early tremors” and expects further disruption as attackers scan the internet for systems that remain vulnerable. That shifts the operational priority from emergency patching to inventory: knowing every place PHP runs, including legacy and forgotten sites, before someone else’s scanner finds them.
AI Works for Both Sides, but Not Symmetrically
The AI framing deserves precision. Humans first discovered these vulnerabilities; AI’s contribution was working out how they could be exploited and validating fixes. That is a narrower claim than “AI found the bug,” but it is the more useful one for defenders, because exploitability analysis is exactly the step that had stalled.
Van der Walt’s caution is about asymmetry. An attacker only has to find a weakness, while a defender has to find a replacement that is, in his words, “patched and robust and tested and approved.” Sullivan put the emphasis elsewhere, calling the episode “quite a decent use case” in which AI surfaced critical vulnerabilities and the community came together to fix them. Both readings fit the facts: AI accelerated a defensive coalition here, and the same accessibility gives attackers more scale and speed.
An Open-Source Capacity Problem Underneath
The PHP Foundation was unusually direct about its own role: “We won’t pretend otherwise: the delay was a structural failure, caused by both the security team’s decision process and its lack of resources.” That candor deserves credit, and it points to a broader issue. Free, open-source software that underpins a large share of the web can depend on security teams with limited capacity to resolve hard engineering trade-offs.
If AI compresses the time between a weakness being understood and being exploitable, the bottleneck moves to the people who design, test and ship fixes. For businesses that run on that software, from hosting providers to cloud platforms, the argument for contributing engineering time or funding to upstream security work gets stronger.
Background
PHP is a free, open-source, general-purpose scripting language built for web development: code that runs on a server to assemble a page before it reaches a visitor’s browser. Its openness has made it ubiquitous. It underpins content management systems such as WordPress and Drupal, which governments commonly use for their sites, and MediaWiki, the software behind Wikipedia. The PHP Foundation works to provide transparency and accountability for the language, and its security team is responsible for developing fixes to vulnerabilities.
The episode lands as AI tools become widely accessible to both security defenders and attackers. Organizations of all sizes can now use AI to find and respond to threats more proactively, while bad actors can use the same capabilities to probe for weaknesses at greater scale and speed. Wiz, a cloud security platform, and OpenAI both said AI played a role in identifying and addressing the PHP flaws. Source: What to know about the AI-detected security bug impacting wide swaths of the internet (AP News) — AP’s explainer on two critical PHP vulnerabilities, the global patching effort and AI’s role in the response.Sources