When half a dozen major services, from Microsoft 365 to the FAA and United Airlines, go down in a single week, it’s natural to suspect a massive, coordinated cyberattack. As of November 18th, 2026 none of these high-profile outages has been attributed to hackers. Almost all of them trace back to a change the organization made itself: a software defect, an internal update, routine maintenance or a new piece of equipment that overloaded the network.
The one confirmed attack of the week told a different story. On November 16, attackers tampered with an Optimizely script that nearly 180 websites load on their pages. They didn’t break into a single one of those companies. They compromised one trusted vendor and let the supply chain do the rest.
What’s Actually Going On
The table above tells two stories.
The first is about fragility. Six outages came from ordinary internal changes, like a bug, a router or an update, and each one still grounded flights, locked people out of their bank or took email offline for millions. That’s how tightly connected our systems have become.
The second story is the Optimizely attack, and it’s the one security teams should pay attention to. It didn’t knock anything offline or set off alarms. It quietly used a trusted vendor to reach nearly 180 websites at once.
The Supply Chain Risk
The Optimizely attackers didn’t need to breach 180 companies. They needed to breach one provider those companies already trusted. Thousands of businesses rely on a relatively small number of infrastructure and technology vendors, and every script, integration and partner portal is a door that someone else holds the key to.
Today, attacks like this still take human time and effort.
Now imagine every one of those steps above run by AI agents that never sleep, test thousands of doors at once and move from first access to full control within hours. That future is closer than it sounds. Attackers are getting faster tools, and they’re also finding more doors to open.
Why the Cracks are Multiplying
Part of the reason is how software is built today. With Vibe Coding, people describe what they want in plain language and AI writes the code. We’re producing far more software than we were a year or two ago, and much of it ships faster than anyone can review it.
That speed comes at a cost to quality, testing and security. As we noted in our 2027 predictions, AI-written code often looks fine and runs fine while nobody fully understands what it misses. We believe all this fast, lightly tested code is leaving more systems behind that nobody watches. Think forgotten integrations, half-finished portals and unmonitored apps. Those are exactly the openings attackers and their AI agents look for.
ARTEX: The Tool the FBI Is Warning About
On November 13, the FBI issued a flash alert on ARTEX, an open-source framework originally built for authorized penetration testing that attackers have turned into an attack tool. An operator gives ARTEX a target and a goal in plain language. AI agents then handle reconnaissance, find vulnerabilities, exploit systems, escalate privileges and steal data, with little human input after that.
The numbers in the FBI alert explain why it matters:
- Speed: Some ARTEX-driven intrusions went from first targeting to privileged access in under three hours.
- Low barrier: Attackers with minimal technical skill can run complex operations against many targets at once.
- Supply chain focus: The FBI says ARTEX is especially good at finding overlooked systems such as partner and supplier portals, mobile backends, legacy services and third-party integrations. It has also been tied to a supply-chain breach at ConnectWise that reached Microsoft 365 customer environments.
- Hard to stop: The original developer pulled the project, but forks, archived copies and a commercial version keep circulating.
Put the pieces together and the picture is clear. Vibe coding is creating more weak spots, supply chains connect those weak spots to thousands of organizations, and ARTEX gives attackers a way to find and exploit them at machine speed. That’s the domino effect we should be preparing for.
Attackers Can’t Evade the Network
Here’s the part that often gets lost in the panic, no matter how fast or autonomous an attack becomes, it still has to cross the network. An AI agent has to look up domains, reach its command servers, move from one system to the next and send stolen data out. A tampered vendor script has to call home. Agentic or not, every attacker shows up on the network somehow, someway.
ARTEX makes that even clearer. Its speed is also its weakness. It works in bursts, checks in on predictable schedules and probes far more than a careful human would. That noise is evidence, and it’s sitting in your network traffic right now.
Attackers will keep getting faster, and the tools will keep changing. What doesn’t change is where they leave their trail. The question isn’t whether an attack will show up on your network. It’s whether you’ll know when it does.
How to Defend Against Autonomous Threats
- Make network visibility a priority: Agentic attacks resolve domains in bursts, check in with their servers on a predictable schedule and scan far more than a careful human would. All of that shows up in network traffic long before the damage is done.
- Don’t trust your vendors blindly. Watch them: The Optimizely attack reached nearly 180 websites without breaching a single one of them. Map every third-party vendor and understand potential risks. A trusted vendor contacting an unknown domain is your earliest warning.
- Find your forgotten systems and unauthorized apps before attackers do. Inventory the apps, portals and integrations built quickly in the last year, especially AI-generated ones.
This week’s outages were a preview of how far one failure can spread. Attackers are moving at machine speed, but so can defenders. Every attack, agentic or not, leaves a trail on the network. Teams that can see it and act on it fast will stay ahead. Lumu gives security teams visibility into every third-party, device and connection, so they can stop internal and external threats at machine speed.