One Forgotten Key Let an AI Hacker Into Barracuda 

Picture of Spencer Thomason

Spencer Thomason

November 18, 2026

Copy Link
One Forgotten Key Let an AI Hacker Into Barracuda

On November 13, 2026, the FBI issued a flash alert about Artex, an open-source hacking framework driven by AI agents. Attackers have been using it since at least August. In some cases, intrusions moved from initial targeting to privileged access in under three hours. Days earlier, security company Barracuda disclosed an attack that matched Airtex. These are real breaches, not lab demos. The FBI, Barracuda, and Amazon have all published what they found.

What Is Artex?

According to the FBI, Artex is a penetration testing framework first released on GitHub in July 2026 for authorized security testing. You give it a goal in plain English, and AI agents turn that goal into commands. It plugs into outside AI models. The FBI says attackers mostly pick low-cost ones with few guardrails.

The developer pulled it from GitHub in October after reports of abuse. Copies and forks are still out there. The FBI says many different attackers use it. One campaign targeted South Korean financial firms and stole personal data on about 67,000 people. Another was a supply chain attack on ConnectWise cloud backup customers that led to sign-in attempts on more than 2,200 downstream tenants. Artex is also good at finding systems nobody watches, such as partner portals and old apps.

How Barracuda Was Compromised

Barracuda sells security. Its security chief laid out exactly what happened. Barracuda gave a cloud access key to an outside AI vendor. When the work ended, nobody retired the key. In Barracuda’s own words, that is on us.

In October, the vendor’s development environment was compromised. The key was sitting in a configuration file in the vendor’s GitHub repository. The attacker did not break in. It logged in.

On November 1, the attacker started using that key. Every secret it found led to the next system: cloud accounts, a data analytics system, source code, and Kubernetes clusters. In roughly 44 hours, it touched about 10 cloud accounts, left a backdoor container behind, and went hunting for crypto wallets and seed phrases.

One system it reached stores the credentials Barracuda’s XDR service uses to connect to customer systems. Barracuda found no direct evidence those were accessed, but its logs could not rule it out. So it assumed the worst and told potentially affected customers to revoke and recreate them.

Barracuda has not said who did it. Many groups use Airtex, so the match shows how, not who. An old key you forgot about is an open door.

What Amazon Saw on Its Decoy Network

Amazon adds another piece. Its decoy network, called Madpot, pretended to be AI model services. From October 3 to November 5, AI agents running Airtex sent requests to those decoys, carrying their own campaign notes.

Amazon says the biggest campaign was stealing keys for AI services. The agents searched for exposed AI tools like LightLM, DeFi, Langflow, and Allama, pulled out the provider keys stored inside, tested them, and pulled them. Then the stolen keys paid for the AI that hunted for more keys.

Amazon cannot verify every outcome the agents reported. It assesses with moderate confidence that several separate China-linked groups ran these campaigns. Amazon says none of the techniques were new. The agents succeeded where basic security hygiene was missing. These agents are not using magic. They use your mistakes.

Steps to Take First

List every key you have given to a vendor, contractor, or tool. If the project is over, revoke it today and use short-lived keys with minimal access. If a vendor no longer needs a key, kill it. Scan your code and build pipelines for secrets and rotate anything you find.

Inventory everything you expose to the internet, including old portals, and put admin pages behind private access. If you run AI tools like LightLM or DeFi, change the default passwords, keep admin pages off the internet, and watch your AI bill. A surprise bill may be the first sign someone stole your key. Be ready to rotate secrets fast because the attack may be over before you notice.

Speed Is the New Threat

The panic version says AI hackers are unstoppable. The evidence does not show that. Amazon says the agents succeeded where the basics were missing. The lazy version says this only hits big companies. Barracuda is a security company and one forgotten key was enough. Attackers can now find your mistakes in hours, not weeks.

StartupHakk Security runs penetration tests and security assessments for companies of every size. A free scan is available at security.startuphack.com to help find the holes, patch what matters, and locate forgotten keys before an AI agent does.

Local Infrastructure and the Democratized Thesis

Why This Matters for Founders

How many critical vulnerabilities are sitting in your production system right now that you have no idea about? Most founders can’t answer that question.

Spencer, founder of StartupHakk, has been building software for 27 years, architecting and scaling platforms for places like GoDaddy, energy companies, and banks. The same fatal mistake appears every day: founders shipping hard, hustling for product-market fit, using AI to write code faster than humanly possible, while security stays on the “we’ll get to it later” list. Developers move too fast and stay too close to the code to see the cracks. Familiarity creates blind spots until one day it isn’t later anymore. One SQL injection, one exposed secret, one broken access control, and the company ends up in the news for all the wrong reasons.

That is why StartupHakk Security was launched. Getting started is easy. Two AI-powered security services are offered. Both are free to start. Both deliver real findings within an hour. Both are engineer-verified so results are not a flood of false positives.

The first is the AI Site Scan. Drop your URL. Non-destructive, security-focused playbooks run against your live attack surface, looking for injection flaws, broken access controls, cross-site scripting, and open redirects. The free tier gives the top three highest-severity findings with actual evidence, not a noise dump.

The next is AI Code Scan Review. Zip your codebase and upload it. No GitHub access is required. Every file is scanned for hard-coded secrets, vulnerable dependencies, injection flaws, and broken access control. Again, the free tier gives the top three highest-severity findings with clear remediation guidelines.

Both features are powered by an AI harness at openmonoagent.ai. This is not placed anywhere in the cloud. It is private control built for actual risk, not a marketing checkbox.

Automated scans are the first line of defense. They catch the low-hanging fruit fast. They do not replace humans who think like an attacker. When ready to move to the next level, engineer-driven deep security review covers full penetration tests, architectural assessments, real attack simulations, and prioritized remediation roadmaps—the kind of work that holds up when money or compliance is on the line.

Security does not have to slow you down. Done right, it can become the reason you can move faster without having to look over your shoulder.

Stop guessing and start knowing. Go to security.startuphakk.com now to run the free AI scan or get a free AI Code Review and see what the AI agents find in less than one hour. If deeper work is needed, book a conversation to start with a human review.

Share this post
Copy Link
Fractional CTO · AI Builds

Stop renting intelligence. Start owning it.

More to explore