AI-Discovered Vulnerability Threatened 70% of Websites; More to Come?

CYRIN Newsletter

AI-Discovered Vulnerability Threatened 70% of Websites; More to Come?

In late October, government websites across Europe began going dark. The scenario cybersecurity professionals have been dreading in this new era of artificial intelligence had finally arrived. A critical flaw was detected in PHP, the programming language used to power websites since the 1990s and that still runs around 70 percent of all websites, including government portals, social media platforms, and other popular sites like Wikipedia. Referred to as PHP2Shell, CISA released a security bulletin that the bug targets two security flaws in PHP — CVE-2026-109618 and CVE-2026-109619 — and Reuters noted that “both rated 9.2 out of 10 on the CVSS v4.0 severity scale, a standard industry scoring system.”

All PHP is at Risk

This high threat level means the execution of unauthenticated remote code, which means a hacker can run commands on the compromised internet server with no password and no account. Voldemaras Kadys, head of security at Mediatech, was quoted by CBS News that these weaknesses “could enable cybercriminals to seize control of unprotected website servers, install malware, pilfer customer data and even take sites down completely.”

According to Wired, these vulnerabilities have impacted all PHP installations as far back as PHP version 5, but only versions PHP 8.2 or higher will receive patches.

This flaw was first observed in August, when researchers at AI firm Calif and at Google’s Wiz discovered the vulnerabilities independently. Using OpenAI’s GPT-5.6 Sol model, “Calif reported the first flaw to PHP on August 11.” Wiz found the flaws on August 24 using its own Wiz Atlas security AI agent, Google’s Gemini 4 Argon, and GPT-5.6 Sol.

NSA Warns Organizations

CNBC reported that “The U.S. National Security Agency quickly warned millions of organizations to implement patches immediately or take their servers offline.” Cybersecurity officials in the EU, UK, and Australia also released warnings. Unfortunately, actions weren’t taken quickly enough. The code claiming to exploit PHP2Shell for any PHP server had already appeared on GitHub and other online forums. As reported by Cloudfare and others, attacks were beginning to emerge and Sansec, a security firm, was predicting “mass attacks beginning within fifteen minutes of the release.”

The most alarming part of this story is how long it took for someone to detect the problem. While these vulnerabilities have been on the radar for years, perhaps decades, exploiting them required skills that very few human researchers could achieve. Two years before, when he was an engineer at Quarkslab, Mihail Kirov, as part of a team, located a “vulnerability with the same root cause as PHP2Shell.” The PHP Foundation was alerted, but a fix wasn’t rolled out — the Foundation has since acknowledged this delay, due to a “structural failure.” However, the basic problem is that the development and proliferation of large language models has changed the landscape. According to Kirov, when using these models, “things can go really, really fast.”

City Governments and Other Agencies at Risk

Matt Suiche, a researcher at cybersecurity startup Tolmo, warned that city governments and other agencies often use PHP and “haven’t updated their websites in a decade.” Even organizations that patch quickly may never know whether they were breached; unless an application security platform was already installed, suspicious requests leave no trace, and as pointed out by Suiche, organizations might not know they’ve been hacked until their “valuable data was being sold later on the dark web.”

What makes this story so important is this: the window of time between discovering a vulnerability and exploitation is shrinking rapidly. Speaking to CBS News, Andrés Guerrero-Saade, vice president of intelligence and security research at SentinelOne, describes how AI is exploiting security flaws more quickly than researchers can detect them, let alone find a solution: “That whole gap between the patch being available and attackers getting to exploit that before anybody patches, that’s the whole game essentially.”

The issue first appeared in April, when Anthropic launched Mythos, an artificial intelligence model so powerful in its ability to exploit software vulnerabilities that Anthropic wouldn’t release it to the public, only making it available to agencies or companies that had been vetted in a cybersecurity initiative called Project Glasswing. Mythos can spot subtle bugs and self-correcting mistakes; in tests it located critical weaknesses in every widely used operating system and web browser.

Impact on Microsoft

According to reports in ProPublica, in the space of one month (April 2026), Mythos discovered “90 critical bugs and 141 important ones in SharePoint, Microsoft’s widely used collaboration software. In June, Microsoft issued patches for some 200 bugs, at the time a historical high, then in July it blew past that mark with patches for more than 600 bugs.

Microsoft products are especially at risk because they are used globally and many of them contain “legacy code” that was developed decades ago. The code carries what is referred to as “technical debt,” or shortcuts and outdated code that can be more readily exploited. In early October, Jamie Dimon from JPMorgan offered a number for this risk on Bloomberg TV, saying “cyber risk went up ten-fold after Mythos,” adding that “AI created vulnerabilities that we didn’t know about, and we always worried about cyber before these things.”

For all the trouble that PHP ignited, this is likely just the beginning. Citrix, BeyondTrust, and others were also impacted, but these stories weren’t as widely covered by the press. The difficulty firms now face is responding to attackers who move at machine speed without the constraints defenders carry. Defenders must test patches, protect operational continuity, and work through legacy systems that cannot be taken offline without affecting production, while attackers need only one crafted upload to any server they can find. As one security consultant, Davi Ottenheimer, put it in Wired, this bug sits in “a file upload interface that is like a front door with no lock” and “an attacker does not need to know which site they are hitting or even what software it runs.”

Market Impact

The shift has impacted the market as well: MarketsandMarkets projects the AI cybersecurity market will grow from “$31.25 billion in 2026 to $95.25 billion by 2031.” This 25% compound annual growth rate is fueled mostly by AI-powered threats, as threat actors are utilizing AI on a greater scale in order to accelerate phishing, reconnaissance, vulnerability discovery, and malware development.”

According to Neatsun Ziv, CEO of cybersecurity provider Ox Security, “The amount of time that it takes the bad guys, the threat actors, to actually find a new exploit and run it on the entire internet used to be weeks,” Ziv said. “It went down to days, and starting in the Mythos age, from April 2026, it went down to hours.”

What PHP2Shell illustrates is this: the contest itself has changed shape, method, and effectiveness. Patching immediately or taking a system offline until it can be patched is now baseline protocol. The essential question is who will get there first: the people finding the problems or the attackers using opensource AI to locate security vulnerabilities. The so-called “good” AI can locate the flaws but finding them is only the beginning. This race will be won or lost in the hours between the patch and the exploit, and that window of time continues to shrink.

Update (November 18 2026): Since this article was written, a similar class of vulnerabilities in Java was discovered, threatening Java-based web applications in the same way that PHP2Shell threatened PHP-based web servers. Fittingly, this new vulnerability was called Java4Shell in a recent CISA warning.

How Can CYRIN Help?

We live in an increasingly dangerous and dynamic world, where cybersecurity flaws can be increased exponentially with the use of AI. It can become our partner, our friend, but in the wrong hands it can be a dangerous adversary. At CYRIN we’ve been working for years on these problems, setting up realistic skills-based training to develop systems and answers to some of these vexing questions.

At CYRIN we understand that continuing innovation is needed as the marketplace continues to change. That’s why we stress continuing education and training, because the job is never done. We continue to work with our industry partners to address major challenges and set up realistic scenarios that allow them to train their teams and prepare new hires for the threats they will face. Government agencies have been using CYRIN for years, training their front-line specialists on the real threats faced on their ever-expanding risk surface.

For educators, we consistently work with colleges and universities both large and small to create realistic training to meet the environment students will encounter when they graduate and enter the workforce. In an increasingly digitized world, training and experiential training are critical. A full-blown cyberattack is not something you can prepare for after it hits.

Our training platform teaches fundamental solutions that integrate actual cyber tools from CYRIN’s labs that allow you to practice 24/7, in the cloud, no special software required. Our new programs, including our new “mini labs,” AI, and Digital Twins, can create real-world conditions for you to practice before you must act. Cyber is a team effort; to see what our team can do for you take a look at our course catalog, or better yet, contact us for further information and your personalized demonstration of CYRIN. Take a test drive and see for yourself!

< Read other CYRIN Newsletters

Contact Us for details or to Set Up a CYRIN Demo
+1-800-850-2170 • sales@cyrintraining.com

Watch CYRIN: The Next-Generation Cyber Range

Learn More About How CYRIN Online Training Can Benefit You