Story · published 26 Oct 2026 · 5 min video
Black October: AI found a 10-year-old hole in the web
For more than ten years, two serious flaws sat inside PHP, the code behind a huge part of the web. In August 2026, AI helped two security teams find them, and PHP shipped the fixes on 21 October 2026.
Watch on YouTube
Key facts
- Two flaws in how PHP handles memory: CVE-2026-109618 and CVE-2026-109619. PHP rated each 9.2 out of 10 (CVSS v4.0).
- Wiz says the flaws were present for more than a decade.
- One of them was spotted in a 2024 audit and not fixed, according to the PHP Foundation (reported by The Next Web).
- A team called Calif used OpenAI's AI and found one flaw (reported 11 August). Working separately, researchers at Wiz used AI and found both (reported 24 August).
- PHP warned users on 20 October and released fixes on 21 October 2026: versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35.
- The flaws could be reached by a single request from someone with no login. PHP 8.1 and older no longer get project updates.
- The same month, Microsoft shipped more than 1,400 security fixes on 13 October.
What to do today
Run a website on PHP? Update to 8.5.12, 8.4.27, 8.3.36 or 8.2.35. On 8.1 or older, talk to your host now.
Use AI agents? Check what they can reach: files, accounts, payment cards. Give each one only what its job needs.
Questions people ask
Did AI really find the PHP flaws?
AI helped two security teams find them. Calif used OpenAI's AI and found one flaw; Wiz used AI and found both. People ran the work, checked the results and reported them to PHP.
Which PHP versions are fixed?
8.5.12, 8.4.27, 8.3.36 and 8.2.35, released on 21 October 2026. PHP 8.1 and older no longer receive project updates, so ask your host about upgrading.
Were websites actually attacked?
Security firms, Cloudflare among them, reported attack attempts after the fixes were published (reported by SC World). We have not seen a confirmed public report of a successful break-in, so we don't claim one.
Why is it called Black October?
Security people used the name for this month because of the number of serious fixes, including PHP's and Microsoft's record Patch Tuesday (Axios, CNBC).
What does this have to do with AI agents?
The same kind of AI that found these flaws can be used by anyone. AI agents that can run commands, change files and use online services need limits, rules that are enforced, a record of what they did, and a human in the loop. That is the LEASH.
Full transcript
The video's narration, word for word, by chapter.
0:00 Black October
Here is something that really surprised me. For more than ten years, a hole sat in the code behind a huge part of the web. One of the flaws was even spotted in 2024, but it was never fixed. This summer, AI helped find them again. Security experts actually have a name for this month: Black October. In the next five minutes, I will show you what happened, what the AI really did, and what it means for you if you use AI agents. I am Ashu, and this is Leashtra.
0:36 Ten years inside PHP
To understand this, let's start with PHP. It is the code language behind a huge number of websites: online shops, blogs, login pages. When you open a web page, there is a good chance PHP made it. Deep inside PHP, there were two small mistakes in how it uses memory. They were there for more than ten years.
0:58 How AI found the flaws
Then, in August, AI helped two security teams find them. A team called Calif used OpenAI's AI and found one of the flaws. Working separately, researchers at a company called Wiz used AI and found both. OpenAI, whose AI helped with the work, said that as far as they know, AI found these flaws before attackers did. Honestly, that is a relief.
1:22 The fix and the race
On October twentieth, the PHP team warned everyone: an urgent fix is coming tomorrow. And this was serious. On the scale that experts use, each of the two flaws scored nine point two out of ten. On the twenty-first, the fixes came out. And then the race began. Because the moment a fix is public, attackers can study it to find out where the hole was. From that moment, every website that has not updated is a target. The news site Axios reported that the NSA, a United States security agency, told website owners: if you cannot update fast, take your site offline. Imagine being told to switch off your own website. Dozens of sites did exactly that. And security companies, including Cloudflare, said attackers were already trying to use the flaws.
2:12 Why it matters
So what can someone actually do with a flaw like this? In simple words: a stranger sends one request to a website, and the website's server runs the stranger's code. Not your code. Theirs. What worries me is this: on some common setups, PHP's own warning says this can happen before the website's own code even starts. So a well-built website was not enough. The ground under it was cracked. Wiz called these some of the most serious flaws ever found in PHP. Strong words, especially coming from the people who found them.
2:50 PHP was not alone
And PHP was not alone. On October thirteenth, Microsoft released more than fourteen hundred security fixes in one month. That is a record. One of those flaws was already being used in real attacks. It was the fourth record month in a row. According to the reports, one reason is that Microsoft now uses AI to check its own software, and the AI keeps finding more.
3:15 How the AI was used
Now here is the part that matters most. The AI that found these holes is the same kind of AI that anyone can use. This time, the good side used it first. But nothing stops an attacker from pointing it at the next piece of software. Which AI was it? Wiz says it used its own AI tool together with AI models from Google and OpenAI. And one thing is clear: the speed. AI now finds holes faster than people can fix them. Think about that for a second.
3:45 The leash
And this is where this channel comes in. An AI agent is an AI that can take actions: run commands, change files, use online services. Give that power to an agent with no limits, and you have built the attacker yourself. So, five things. Limit: give the agent only what it needs. Enforce: put the rules in the tool, not just in the prompt. Audit: record every action it takes. Secure: keep your real passwords and keys away from it. And Human: a person says yes before anything big.
4:19 What to do today
If your website runs on PHP, update it today to one of the versions on screen. If you are on PHP 8.1 or older, the PHP team no longer sends fixes for it, so talk to your hosting company now. And if you use AI agents, check what they can reach. Today, not next week.
4:36 Closing and what's next
Black October is not the end of something. It is the start. AI is now part of security on both sides, and the winners will be the people who keep their AI on a leash. All the sources are in the description. And coming soon on this channel: your AI agents do real work all day, and the next morning, they forget all of it. We built something to fix that. It's called WWW. Every AI wants to be an Astra. We give it a Leash.
Sources
- PHP security advisory and releases, 21 Oct 2026 (primary): php.net
- Wiz Research, "PHP2Shell", 21 Oct 2026 (primary): wiz.io
- OpenAI, "Patch the Planet", on the Calif and Wiz findings (primary): openai.com
- Axios, "Black October", via Yahoo News, 22 Oct 2026: yahoo.com
- PHP Foundation on the 2024 audit, via The Next Web: thenextweb.com
- SC World, attack attempts reported by Cloudflare: scworld.com
- BleepingComputer, Microsoft October 2026 Patch Tuesday: bleepingcomputer.com
- Redmondmag, another record Patch Tuesday: redmondmag.com
- CNBC, 23 Oct 2026: cnbc.com
How this was made: the host, Ashu, appears as an AI avatar, and the voice is an AI copy of Ashu's voice. Every fact on this page was checked against the sources above. Last checked 25 Oct 2026.