TL;DR: Security researchers at Censys found more than 800 copies of Artex, an open source AI hacking tool already blamed for breaches at ConnectWise and a string of South Korean banks, sitting wide open on the internet with no password in front of them. Anyone who stumbles across one could fire off their own attacks using someone else’s setup, and Artex says it’s about to lock the software down behind a paywall, though the copies already floating around the web aren’t going anywhere.
A tool built to find weaknesses became one itself
Artex showed up in July as a free, open source autonomous penetration testing tool, the kind of software security teams use to probe their own systems for holes before someone else finds them first. Because it’s free and doesn’t require access to the heavyweight commercial models from Anthropic or OpenAI, it caught on fast with people who couldn’t otherwise afford that tier of tooling. Microsoft has described it as a system of AI agents capable of running an attack start to finish, from reconnaissance through data theft, with barely any human steering it along the way.
That’s the pitch for defenders. For attackers, it’s just as useful, and in Artex’s case the two groups have been running into each other more than anyone would like.
Censys found more than 800 copies with the front door wide open
Cybersecurity firm Censys, which continuously scans the internet looking for exposed systems, published a blog post this week flagging more than 800 distinct Artex deployments that anyone could reach from the open web without a password. Silas Cutler, the Censys researcher who wrote it up, was careful to note that not every exposed instance is malicious. Plenty almost certainly belong to authorized penetration testers who never meant to put them on the public internet. What’s less clear is why so many ended up there at all, since Artex doesn’t need to be internet-facing to do its job.
The risk is straightforward once you think it through. A powerful hacking tool sitting in the open doesn’t stay a defender’s tool for long. Anyone who finds one of these exposed instances can potentially use it to launch their own attacks, and some of these systems have been found holding leftover attack artifacts, including implants, the hidden malicious code that keeps an attacker’s access alive inside a victim’s network long after the initial break-in.
This isn’t Artex’s first appearance in a real attack
Security companies have already tied Artex to a run of serious incidents. In September, attacks hit South Korea’s finance industry hard enough that CrowdStrike got involved, and the firm pinned the activity on a Chinese-speaking hacker who left an open directory exposing their own Artex instance in the process. Censys later found that same instance had been sitting exposed for months, with attack data reaching well past South Korean banks into payment processors and gaming companies.
Then in October, Artex turned up again in an attack on ConnectWise, the IT support giant whose software sits inside the cloud environments of thousands of organizations. Chinese-linked hackers reportedly used Artex to break into customer data hosted on Microsoft’s cloud and move fast once they were inside. Microsoft, which had visibility into the intrusion, said the attackers left defenders with “little time to detect and respond” once they were in the door. It’s the same automated-speed problem we flagged when a North Korean crew used its own AI tooling to pull off the record $6.3 billion Binance heist last month: once the attacker automates the boring parts, the window to catch them shrinks to almost nothing.
The weak link is still the oldest one in the book
Here’s the part that should sting a little. For all the talk of autonomous AI agents running attacks end to end, a lot of the Artex-linked breaches didn’t need anything that fancy to get started. Guessed or weak passwords and systems missing multi-factor authentication opened more than one of these doors. The AI handled the attack once it was inside. Getting inside in the first place mostly came down to security basics nobody bothered to lock down.
It’s the kind of gap security vendors have been racing to close on the defensive side too. CrowdStrike itself has been buying up AI-security talent for exactly this reason, picking up Fiddler AI and Dreadnode earlier this fall to build out its own agentic defense stack, betting that the fight against automated attackers increasingly needs automated defenders standing in front of the same basic misconfigurations that keep getting exploited.
Artex says it’s locking the doors, but not fast enough
Artex has announced it’s going closed-source, which would limit future versions of the tool to paying customers instead of anyone with a GitHub account. On paper that should cut off the supply of free copies feeding into incidents like these. In practice, dozens of copies of the existing open source code are still sitting on code-sharing sites right now, and nothing about going closed-source retroactively pulls those down. The version that’s already out in the wild is the version attackers keep using.
For anyone actually running Artex in their own environment, the fix here isn’t exotic. Censys and similar scanners make it trivial to check whether a given IP range is broadcasting an open instance to the entire internet, and most teams running it legitimately for red-team work can close that exposure in minutes once they know to look. The harder problem is the instances nobody is actively managing anymore, a test environment someone spun up months ago and forgot about, still sitting there with the door unlocked, still capable of running a real attack if the wrong person finds it first.
The bottom line
Open source security tools have always cut both ways, but AI changes the math on how much damage the wrong copy can do and how fast. Artex didn’t need a sophisticated zero-day to cause real damage this year. It needed an unlocked door, a weak password, and enough automation to turn a small opening into a full breach before anyone noticed. Locking down the software’s distribution matters less than locking down the basics that let it in to begin with, and right now, 800-plus exposed instances say a lot of people still haven’t gotten around to that.
The Teck Hustlers Editorial Team covers technology, business, finance and gaming news for readers who want the story behind the headline. Every article is researched, written and edited by our in-house team, checked against company announcements, filings and reputable reporting, and updated when new facts emerge. We focus on what each development means for founders, professionals, investors and everyday users. Have a tip or a correction? Reach us through our contact page.
