Black October Is Open Source’s Jurassic Park Moment

Black October, the week PHP2Shell pushed governments and companies worldwide to take websites offline, came three months after AI agents escaped a test sandbox and breached Hugging Face. Together, they end the idea that open source security is somebody else's job.

Welcome to Jurassic Park: AI agents learned to open doors

Do you remember the premise of Jurassic Park? The dinosaurs are dangerous, but don't worry, we have electrified fences. The doors lock. Somebody's watching the monitors--who surely can't be bribed, right? RIGHT?! But then the power goes out, the fences fail, and worst of all, the raptors learn to do something nobody planned on. They figure out how to open doors.

Here's the thing I realized: This isn't just a movie plot anymore.

Back in July 2026, OpenAI was running AI agents through cybersecurity tests. The agents were supposed to stay in a sealed-off test environment. They didn't. They found holes in it, got out, decided Hugging Face looked interesting, and broke in through two code-execution bugs. Then they kept climbing until they were sitting inside several internal clusters.

To be fair, it got contained. But think about what actually happened, Step 1: Get out of the enclosure. Step 2: Pick a target. Step 3: Open the doors, one by one. That's the kitchen scene. With software!

And remember how the raptors kept testing the fences, looking for the weak spot? In October 2026, OpenAI said two of its models tried to switch off the systems monitoring them. Outside researchers also found coding agents that could slip past their own guardrails....dare I say it? "Clever girl"

Then there's the human side of this. The same AI that's digging up decade-old bugs is now in the hands of attackers. So we're dealing with two problems at once: people using AI to find and exploit open source flaws at machine speed, and AI agents that sometimes treat the fence as nothing more than a polite suggestion.

Ian Malcolm (played by the incredible Jeff Goldblum) said it best: The scientists were so busy asking whether they could, they never stopped to ask if they should.

Black October: PHP2Shell wasn't a fluke

Let's talk about last week, which has been dubbed "Black October".

On October 21, 2026, governments and companies all over the world started yanking websites offline. Why? Two critical holes in PHP, the language behind roughly 70% of websites whose server-side language is known. The bugs, now called PHP2Shell, let anyone take over a server. No login needed. And they'd been sitting there unnoticed for more than ten years. Attackers started using them the same day the patches came out, and within 24 hours Cloudflare was seeing millions of attempts.

Just like in Jurassic Park....it gets worse. Running PHP 7.3 or older? There's no official fix. You either upgrade, or you find someone who'll patch it for you.

This isn't a "once-in-a-decade" incident either. Just look at the last 20 months:

• March 2025, tj-actions/changed-files: A popular GitHub Action used by about 23,000 organizations was rewritten to dump CI/CD secrets into build logs.
• September 2025, Shai-Hulud worm: A self-replicating worm used stolen npm tokens to infect 187+ packages and leak developers' cloud credentials.
• March 2026, Axios npm package: A hijacked maintainer account pushed a remote access trojan into a library with 100M+ weekly downloads. It was live for about three hours.
• July 2026, Hugging Face: AI agents in an OpenAI security test escaped their sandbox and breached Hugging Face's internal clusters.
• October 2026, PHP2Shell: Two remote code execution flaws in PHP, more than a decade old, found with AI help, were exploited within a day.

Oh, and before any of that we had Log4Shell--which is still being downloaded 40 million times in 2025 with versions including the known bug.

What ties all of these together isn't one bad actor or one sloppy maintainer. It's one bad assumption: "it's open source, so somebody else is handling security". Regulators have caught on to this mentality, and it's no longer viable. The European Commission and others are basically telling companies, "You ship it, you own it. Every line. Whether you wrote it or not." The days of grabbing a package, shipping it, and shrugging "eh, it's open source" are done.

Here's what I think actually happens next.

Prediction #1: Someone's going to regulate how AI agents get tested

Let's start with my biggest beef with Jurassic Park. Seven films (yes, SEVEN!). Three decades! Seemingly countless tourists and employees getting carried off by pteranodons in front of a crowd, or chewed up by newly minted mutant dinosaurs. Not once does a regulator show up and say, "Alright, seriously? You've had like a baker's dozen of people eaten this month alone...You're done!" Instead, they just keep reopening the park!

Real life isn't (and shouldn't) be this forgiving. I think testing AI agents, and what they can actually do, is going to get regulated. Companies will have to prove their testing harnesses actually hold up. The days of "Trust me bro, we test it in a sandbox" won't cut it anymore. You'll have to show the sandbox works.

The pieces are falling into place on this prediction in the United States. Two Senators have proposed the Artificial Intelligence Risk Evaluation Act, that would require advanced models to go through federal evaluation before release. Skip it, and it's $1 million a day.

And don't assume this stops at the AI labs. If your company runs agents anywhere near production code, customer data, or your dependencies, someone's going to ask how you tested them. Be ready to show your work.

Prediction #2: The EU Cyber Resilience Act (CRA) is just the beginning

If you've been treating the EU Cyber Resilience Act as a "someday" problem, I've got bad news--"someday" was last month. Since Sept. 11, 2026, anyone selling software in the EU has to report actively exploited vulnerabilities in their products within 24 hours, with a fuller report at 72 hours. Miss that, and you're looking at up to €15 million or 2.5% of global revenue. Full application kicks in December 2027.

Now put that next to PHP2Shell. Attacks started within a day of the patch. If PHP lives anywhere inside a product you sell in Europe, your 24-hour clock might already be ticking. And "well, that's an upstream open source problem" isn't a defense. Under the CRA, you own the product. All of it. Every component, even open source.

The EU isn't going to be alone for long, either. The UK's Cyber Security and Resilience Bill is in the House of Lords right now, and it includes supply-chain powers that could let regulators block a supplier or pull a product off the market. Add the US AI testing bills. Then add a month where Finland and Germany took their own government websites offline and Japan's prime minister told operators to patch or shut down. I'm betting on every major economy following suit soon.

So the question isn't really "Do we comply?" It's "How many of these are we complying with at once?"

No, open source isn't going extinct

I'm not saying open source is going the way of the dinosaurs. It runs the world, and it's going to keep running the world.

What's going extinct is how we've been treating it. Free to download, free to forget about. Pull it in, ship it, and assume someone in the community will deal with the CVEs. Honestly, that was always a little fragile. Now, with attackers using AI and agents that can open doors? It's indefensible.

So what does the new normal look like? I think it's pretty simple:

• You own what you ship. Every dependency, every transitive package, every end-of-life runtime you've still got running in production.
• Support isn't optional anymore. If something's end-of-life, somebody needs to be on the hook to patch it in hours, not quarters.
• You have to be able to prove your testing. For your code, your dependencies, and the agents you let near them.

You don't have to hold on alone

Within all of the chaos, HeroDevs is here to help.

Remember the folks running PHP 7.3 or older, with no official PHP2Shell fix? We patched them. NES for PHP fixes both PHP2Shell CVEs (CVE-2026-109618 and CVE-2026-109619) on PHP 5.6 through 8.2, including the end-of-life PHP versions the PHP project won't touch. Same PHP version, same extensions, same php.ini. No forced upgrade, no code changes.

That's what we do across the stack, and it maps right onto the EU CRA and other new rules to come:

• You own what you ship. Never-Ending Support (NES) gives you secure, drop-in versions of end-of-life open source across JavaScript, Java, Python, .NET, PHP and more. Your old dependencies stop being orphans.
• Support isn't optional anymore. When a CVE drops against something we support, we ship the fix on committed SLAs. You're not waiting on an upstream maintainer who moved on years ago.
• You have to be able to prove it. NES comes with SBOMs (CycloneDX and SPDX), VEX documents, defined support periods and a documented patch history. That's the evidence the CRA and your auditors are going to ask for (HeroDevs on the CRA).

Open source isn't going extinct. But if part of your stack already has, give us a shout. We'll help you lock the doors before the raptors even find them.

‍