The risk of exposure is enormous, multiplied by artificial intelligence—but time-tested global cybersecurity procedures appear to be working. Then suddenly, everything fell apart.
A cybersecurity firm’s decision on Wednesday to prematurely release information about a flaw in the popular PHP computer programming language left much of the web open to hackers, igniting panic and fueling months of warnings about the potential for widespread damage in the age of artificial intelligence.
The flaw in PHP was discovered using artificial intelligence tools during a months-long process coordinated by OpenAI, leading online agencies in Europe, the United States, the United Kingdom, Australia and Canada to issue emergency warnings on Wednesday. Many governments have taken their websites offline as a precautionary measure.
While no major hacking damage was immediately apparent, the events unfolded within 72 hours, demonstrating how easily cybersecurity processes can break down and the dramatic consequences of cyber risks in the age of artificial intelligence.
The man who sparked the ripple effect of the early announcement — Chainguard CEO Dan Lorenc — quickly apologized in a LinkedIn post. But on Thursday, Lorenc told POLITICO in an interview that he shared the information in advance because the flaw had already been exploited. While others have disputed this, Lorenc said he believes he has seen enough evidence to invalidate the embargo to allow time to find a way to fix the flaws.
While Lorenc denied that his actions caused problems for other defenders, he said the evidentiary threshold for these coordinated cyber defense systems was a “grey area.”
The revelation of the PHP flaw brings to light what cyber officials have been waiting for since the advent of the most powerful artificial intelligence tools that can be used not only to find vulnerabilities but to exploit them at scale.
Long-established protocols in the cybersecurity world help prevent catastrophic scenarios in which hackers can successfully exploit these vulnerabilities to take control of numerous websites and access their data. But the scale and apocalyptic potential of AI-driven cyber risks are now visible to the world.
“This is what happens when artificial intelligence discovers flaws that have been hidden for years,” said Eylam Milner, co-founder and chief technology officer of Echo, a cybersecurity startup. He added that these vulnerabilities are “the most serious for the Internet.”
Origin story
Understanding how this week’s emergency unfolded can shed light on how companies and authorities are trying to avoid attacks, and how much risk they face in the future.
The origins of the incident can be traced back to April, when Anthropic, a leading cutting-edge artificial intelligence company, launched Mythos, leading some in the cybersecurity community to worry that systems would soon be overwhelmed by advanced artificial intelligence and potentially catastrophic events could occur.
But while researchers did find more flaws in computer systems in the months after other artificial intelligence companies like Anthropic and OpenAI released these models, their real-world impact was minimal. Even the “facehugging” incident—a widely discussed hack that caused OpenAI’s autonomous artificial intelligence agents to spiral out of control during an internal security review—had more to do with the degree of control AI companies have over their models than their ability to break into global systems.
That changed this week after researchers published a vulnerability in PHP, a coding language estimated to be integrated into about 70% of the web.
The process of disseminating information about flaws among cybersecurity officials and affiliates is critical. Through a long-standing practice known as coordinated disclosure, researchers and companies that discover flaws can find fixes and share them confidentially with agencies and others before releasing them to the public, meaning they don’t give malicious hackers the gift of unfixed vulnerabilities.
In the case of PHP flaws, the process works well, but it’s not perfect. Information about the error began to spread before the scheduled release time. Then, according to the PHP Foundation, Chainguard “unilaterally broke the embargo” and “ignored weeks of coordinated disclosure efforts by multiple parties.” Elizabeth Barron, executive director of the PHP Foundation, told POLITICO via email that while it is “disappointed” with Chainguard, it hopes to “rebuild trust with them in the future to ensure we can continue to work together effectively.”
Chaos quickly followed the leak, which meant that information about the vulnerability became public before a fix was ready, with government agencies and private companies choosing to temporarily shut down their sites rather than risk potentially damaging damage to the data and the sites themselves.
Earlier Thursday, Cloudflare, a web services company widely used by websites to thwart cyberattacks, giving the company a high level of visibility into hackers’ behavior, said in a post on X that it had seen “millions” of attempts to exploit the flaw.
Cloudflare discovered thousands of vulnerabilities in the 10 hours before they became public, then tens of thousands within 45 minutes of the announcement, hundreds of thousands within two hours, and now millions, Prince said.
Amazon’s cloud division also said the vulnerability “is now being widely exploited.”
A senior U.S. government cyber official said federal authorities have been overwhelmed, and not just PHP. On the same day that the world is racing to fix problems with PHP, they are also dealing with bugs in widely used JavaScript frameworks Node.js, Fortinet, GitHub, and Oracle.
“The fire department is fighting multiple fires at the same time,” the official said, adding: “The good news is that the fires are all single-story buildings and the fire department is dealing with these fires every day.”
More coming soon
Alarmingly, OpenAI says there are more features to come. “These bugs are among several significant vulnerabilities we have discovered and are working to fix, and we expect similar discoveries to increase dramatically in the coming months,” the company said in a statement Wednesday.
Matthijs van Amelsfort, head of the Dutch Cyber Security Agency, said in an interview that advances in artificial intelligence have changed the work of maintaining computer system security. “As soon as the leading-edge models come out, that’s exactly what we’re saying: This is going to be the trend,” he said. “In fact, that’s happening right now.”
Thomas Caspers, vice president of German internet agency BSI, agreed: “We expect [to be] This will happen more and more often in the coming months. “
Artificial intelligence also increases risks once flaws become public. Caspers said online agencies recommending organizations take their websites offline “is really a drastic measure.” “We wouldn’t have done this a few years ago.” But as artificial intelligence makes exploiting flaws faster and easier, “we have to do it,” he said.
At the same time, using advanced artificial intelligence to detect flaws may provide a new level of protection against malicious hackers trying to exploit these flaws. Kaspers said that although his German agency also discovered the attack, it had not seen hackers successfully exploit the vulnerability.
“It seems to be a little more complicated to exploit than expected,” he said.
While artificial intelligence offers a new way to find flaws, the work involved in protecting systems is more routine, with cyber agencies urging organizations to install fixes shared by the PHP Foundation, a process called patching.
“Now you have to do three things,” said Dutch Agency’s van Amelsfoort. “That’s patch, patch, patch.”
John Sakellariadis contributed to this report.
Axel Springer’s global network of journalists draws on the resources of the company’s newsrooms to publish ambitious scoops, investigations, interviews, opinion pieces and analysis. It allows journalists – including those from POLITICO, The Telegraph, Business Insider, Le Monde, Bild, Onet and Fakt – to collaborate across platforms (online, print, TV and audio) to write breaking stories for hundreds of millions of international audiences.