As the use of artificial intelligence (both benign and adversarial) increases at breakneck speed, more instances of potentially harmful reactions are being discovered.
Pix Deluxe | E+ | Getty Images
European government websites were taken offline on Wednesday amid widespread concerns about vulnerabilities in servers hosting sites coded in PHP, the world’s most popular server programming language.
The US National Security Agency warned millions of organizations to apply patches immediately or take their servers offline. This is the latest cyber-horror in the age of artificial intelligence, where vulnerabilities that once took hackers weeks or months to discover can now be exploited in hours.
Matt Swish, a researcher at cybersecurity startup Tormo, said this level of urgency is “very new.”
“That’s something that didn’t exist a year ago,” he said. “The reason there is such a sense of crisis right now is because everyone has AI-enhanced capabilities, not just defenders but also attackers.”
what happened?
Created in the 1990s, PHP is used by just under 70% of all websites using known server-side programming languages.
In mid-August, researchers at the AI company California Google Wiz has identified vulnerabilities in all PHP installations that handle web requests. For this discovery, California researchers used OpenAI GPT-5.6 Sol, while the Wiz team used the Wiz Atlas security AI agent, Google’s Gemini 4 Argon, and GPT-5.6 Sol.
The vulnerability, known as PHP2Shell, allows an unauthenticated hacker to remotely execute code on any PHP server, potentially allowing malicious parties to steal data or read sensitive information.
The PHP Foundation was scheduled to release a patch for the bug on Thursday. However, news about the vulnerability was first leaked in an online forum, and then by the state of California itself, which accidentally posted a proof-of-concept video demonstrating the exploit chain on Tuesday.
The video was eventually removed, but a patch was quickly tracked down and released by the PHP Foundation 24 hours early at 10 a.m. ET on Wednesday. Shortly before that, the NSA issued a warning.
At that point, the code had already been posted to the public repository GitHub. microsoftclaims that PHP2Shell can be exploited against any server.
Elizabeth Barron, executive director of the PHP Foundation, told CNBC on Wednesday afternoon that the foundation is “closely monitoring the situation,” but that there is currently “no evidence of any working exploits, just incomplete, misguided, non-functional generation.”
However, Suiche said Tolmo’s team was able to code a working exploit in less than a day.
“With AI, we could reproduce it locally relatively quickly,” he said, adding that before today’s powerful AI models, it would have taken his team “weeks” to create it.
Amazon Threat Intelligence reported Wednesday that it identified a PHP2Shell exploit attempt on Saturday on the MadPot telemetry network, a network dedicated to identifying digital decoy connection attempts.
“Structural defects”
This vulnerability has existed for a long time. Two years ago, Mihail Kirov, then an engineer at Quarkslab and now a product security engineer at Tolmo, said he was part of a team that discovered a vulnerability with the same root cause as PHP2Shell.
At the time, Kirov said, the PHP Foundation warned of flaws in the system, but no fixes were published. The foundation acknowledged in a blog post Wednesday that the delay was a “structural flaw caused by both the security team’s decision-making process and lack of resources.”
That was before large language models were so popular. Researchers in California had a powerful new technology at their fingertips. They found two vulnerabilities rated 9.2/10 in terms of security severity.
“They have proven today, in 2026, that by using large-scale language models, things can move very quickly,” Kirov said of the California team. “One of the reasons is that the LLM allows you to explore from so many different angles.”
Even if organizations patch vulnerabilities, they likely won’t know whether their data has been compromised, experts said.
Nietsun Jibu, CEO of cybersecurity provider OxSecurity, said organizations would not be able to see suspicious requests made on their servers unless they pre-installed an application security platform before attempting a hack.
Swish said companies would only realize they had been hacked if their valuable data was “later sold on the dark web.” That was “yesterday’s reality and today’s reality,” he said.
Because PHP is a legacy server language, many of its users may not have implemented advanced security measures, Suiche said.
He said city governments and other agencies often use PHP, but “haven’t updated their websites in 10 years.”
“Those people have a lot of important data,” he said.
Industry experts say the cybersecurity game has changed dramatically since Anthropic released its powerful Mythos model earlier this year. Claude’s creators pre-released Mythos to a small number of businesses and government agencies in April to ensure that cybersecurity measures were adequate before a large-scale release.
Mythos is currently available to many bad actors.
“It used to take weeks for the bad guys, the threat actors, to actually find a new exploit and run it across the internet,” Ziv said. “It has been reduced to days, but since the time of Mythos, it has been reduced to hours from April 2026.”
Attention: The arrival of the era of AI-powered cyberattacks
