This could have been a massive mess.
By Wolf Richter for WOLF STREET.
Wednesday morning, October 21, at a little after 5 AM Pacific Time, as I was starting to read through the business and finance news and reports, I got a jolt from something bigger than caffeine.
Researchers, using the frontier AI model GPT-5.6 Sol, had discovered two remote code execution vulnerabilities in the PHP engine which had been there for many years. These vulnerabilities, CVE-2026-109618 and CVE-2026-109619, collectively known as PHP2Shell, would allow hackers to take over the entire website and do whatever.
The researchers discovered the vulnerabilities in August and were working with PHP, OpenAI, and WordPress to come up with a way to patch the vulnerabilities before announcing it publicly. The date of the release of the patch was supposed to be October 22.
The problem yesterday morning (October 21) was that those vulnerabilities had been leaked into the wild days earlier, but there was no patch yet, and they cropped up in the social media amid reports of attempts to exploit them. This triggered an emergency announcement by PHP at about 7 AM Pacific Time, along with an emergency release of the patch.
PHP is the open-source scripting language that about 70% of the world’s websites use, including all WordPress sites, including WOLF STREET. And there was still no patch when I first saw this. It sent the world ascrambling. The recommendation was to shut down websites running PHP until a patch would be available and was installed.
“These vulnerabilities allow an attacker to send a crafted set of queries to a server running PHP and potentially take over the server. From there they could potentially read, modify, delete, or repurpose anything on the server,” according to Cloudflare, which provides a cloud-based Web Application Firewall (WAF) to websites, including WOLF STREET.
Cloudflare said that it had developed ways to mitigate the PHP2Shell vulnerabilities in the hours before the public announcements.
In terms of the numbers, Cloudflare CEO Matthew Prince said on X last night:
“In the 10 hours before it was publicly announced, across the @Cloudflare network we saw thousands of attempts to exploit it. That jumped to tens of thousands in the 45 minutes post announcement, hundreds of thousands within 2 hours, and millions now.”
Lots of websites in Europe – websites by governments, hospitals, and other essential service providers – had already shut down by the time I looked at this yesterday morning.
In the US, most prominently for WOLF STREET readers, the St. Louis Fed had shut down its data website FRED, that hosts government and private-sector data from all over the US and the world. Lots of our readers use FRED frequently to get their favorite data sets. FRED is a huge site and gets lots of global traffic and it could have become a target.
Amazon’s AWS reported yesterday after the public disclosure of the vulnerabilities:
“Amazon Threat Intelligence identified a PHP2Shell exploitation attempt dating to Oct. 17, 2026, in MadPot telemetry, four days before public disclosure of CVE-2026-109618 and CVE-2026-109619.
“The exploit attempt originated from an IP address associated with a commercial anonymization network. While the vulnerability was under embargo, the targeting of a honeypot from an anonymized IP address suggests the activity was conducted by a threat actor, though it remains unclear how the actor acquired exploit information prior to disclosure.
“The PHP embargo window was set to be lifted on 2026-10-21 1400 UTC [7 AM Pacific Time]. On the day of the release, Amazon Threat Intelligence identified multiple proof of concept exploits for the vulnerability, with the earliest around 2026-10-21 0200 UTC [the evening of Oct. 20 Pacific Time].”
And AWS added:
“Once the embargo was lifted and the patch was released [at 7 AM Pacific Time], additional proof-of-concept exploits appeared within hours, and we saw them used against our sensors almost immediately. PHP2Shell is now being broadly exploited in the wild.”
And it’s not an easy quick patch either. Nearly 200 files in PHP needed to be fixed, with developers writing over 10,000 lines of code, according to the WSJ last night.
So this was not a great thing to wake up to for website owners, for sure. At the time, shutting down PHP sites was the recommendation. I didn’t shut down WOLF STREET, hoping for the best, and counting on the mitigation efforts by Cloudflare and Wordfence (a security plugin). Hope is not a strategy, but sometimes it works.
For Liquid Web, the data center company that hosts the dedicated WOLF STREET server and a gazillion other dedicated servers and Virtual Private Servers (VPS), it was an all-hands-on-deck moment, I’m sure. They scrambled to apply the emergency patch, once it became available, to the most recent versions of PHP and deploy them across their network. PHP was patched on WOLF STREET yesterday afternoon.
Enjoy reading WOLF STREET and want to support it? You can donate. I appreciate it immensely. Click on the mug to find out how:
In this case AI looks more like the firefighter than the Terminator.
The vulnerability was already sitting there in PHP for years. The AI didn’t invent the flaw, it helped researchers find it before widespread criminal exploitation. The discovery led to patches and warnings being issued to website operators around the world.
What’s ironic is that this is probably a glimpse of the future-
AI finds bugs faster.
AI helps write fixes faster.
AI helps defenders scan systems faster.
AI also helps attackers find targets faster.
So it’s less “AI saves humanity” and more:
Humanity just handed both the cops and the robbers Formula 1 cars.
In this round, the cops found the problem first.
The funny headline could be: “AI discovers catastrophic internet flaw, alerts humans, humans immediately panic that AI might also tell the bad guys.”
Based on the reporting – this is pretty close to what happened.
Oh wow, that explains why some of my favourite stat sites were down yesterday, I was getting thoroughly annoyed by it 😆.
I’m starting to think the most potent application of “AI” or “SI” or whatever will be (if they aren’t already) in cyber warfare/security space, afterall the internet is the incubator of this technology.
MW: Nasdaq and S&P 500 sliding in final hour as AI angst returns to the stock market
“But wait! There’s more!”
Axios:
“Major U.S. banks are detecting unauthorized third-party AI agents operating inside customer accounts, with some agents attempting to take actions that appear to be independent of user instructions, sources tell Axios.”
Makes me want some **itcoin. Right?
‘Cause it’s digital so it’s “safe”.
Is AI still a conspiracy theory now that it’s actually happening?
Waiono – I was tech team leader of the 4th U.S. bank on the Internet, and I agree that everything digital is just a wisp in the wind.
Your account at the bank or brokerage is made of pixels as well. One can hope they have hard file backup under a mountain somewhere.
I think they’re talking about bank customers using AI agents to do stuff in their bank accounts, but those AI agents weren’t authorized by the bank. And some of those bank customers’ AI agents were doing stuff that the bank customer might not know about or want.
Don’t ever let an AI agent get into your bank account or brokerage account, keep your passwords away from the AI agent, and if they misbehave, shoot them?
Some people cannot be helped.
It would really be interesting to find out how this information was leaked, who leaked it and are they criminally accountable.
So now the question is if it was actually leaked, which wouldn’t surprise me considering the number of people involved, or if it was found by another party via a similar method.
Well, I guess that explains why I was banned from this website.
If you are afraid of all the bank account bot stories and you live in California, Cashier’s checks have a “use by” date of at least a year at all my banks. I think it is typically either 60 days or 90 days for other states.