Artex Breached in Three Hours: How AI Hackers Outpace Human Reaction Speed
The FBI warns that Artex can connect to multiple large models to autonomously scan for and exploit common vulnerabilities such as weak passwords, compromising systems in under three hours in some attacks; it has already affected South Korean financial institutions, ConnectWise customers, and Barracuda. Even after the developer shut down the old open-source code, unrevoked credentials and exposed instances keep the risk alive.

- Artex can complete an intrusion in under three hours
- At least 2,250 customers' cloud environments affected
- At least 68,000 people's data stolen in South Korea
- Hundreds of Artex systems exposed without passwords
The FBI warns that the open-source AI hacking tool Artex has been used in intrusions at South Korean financial institutions, ConnectWise customers, and Barracuda, with some attacks going from system lock-on to breach in under three hours. At least 68,000 people's personal data was stolen in South Korea, and at least 2,250 ConnectWise customers' Microsoft 365 environments were accessed. The real urgency is not the notion that "AI can hack" — it is that human response speed can no longer keep up with a machine's continuous actions.
In Under Three Hours, Artex Breached Three Networks
The damage first appeared in South Korea. In its latest cybersecurity advisory, the FBI said multiple South Korean financial institutions were hit by Artex attacks, with at least 68,000 people's personal data stolen. The advisory partly attributed the problem to weak security protections and a lack of multi-factor authentication (confirming identity with more than one credential).
ConnectWise has confirmed it was breached. The U.S. IT management company was compromised by hackers using Artex, through which attackers reached the Microsoft 365 environments (Microsoft's enterprise cloud service environment) of at least 2,250 customers. The FBI confirmed Microsoft's earlier findings; Microsoft said affected customers' data was stolen. A ConnectWise spokesperson did not respond to TechCrunch's request for comment on the FBI's statement.
Barracuda's entry point was an old, unrevoked credential. Cybersecurity company Barracuda disclosed that a group of profit-motivated criminals launched related attacks via Artex; the tool discovered exposed credentials for a third-party service, and that credential belonged to a vendor whose partnership with Barracuda had ended but whose credentials had not been revoked.
Attackers used it to reach a storage location holding credentials, which was used to collect data from Barracuda customer systems. Because logs were not detailed enough, Barracuda could not rule out that data was accessed, and ultimately replaced all customer credentials.
The entry points of the three incidents differ, and so do the remediation priorities: where credentials were not revoked, check logs and rotate credentials first; where protections are weak, shore up authentication first.
Artex Compresses Intrusions Into a Three-Hour Window
What Artex changes is attack speed: from target discovery to completed intrusion in as little as three hours, often too fast for defenders to respond.
The U.S. Federal Bureau of Investigation (FBI) says Artex was originally an open-source AI penetration testing platform (simulating attacks to test network defenses). Hackers can connect it to hosted frontier large language models (AI models that can understand and generate text) from companies such as Anthropic and OpenAI, or to open-source and non-U.S. models. The models provide judgment and execution capability, while Artex breaks the attack into a continuous sequence of steps.
The first step is to rapidly scan the victim's network for weak points. The entry points the FBI lists include insufficient protections, exposed passwords, and passwords that have already been stolen. Attackers do not need to crack sophisticated vulnerabilities; holding a single entry point is enough for the automated tool to keep acting.
After gaining an initial foothold, Artex continues to choose subsequent actions, spreading inside the network and searching for databases, mailboxes, and sensitive information. The attack chain does not necessarily depend on some mysterious vulnerability; ordinary credentials and basic protection gaps can also be strung together into an intrusion path.
Once humans take over, the pace slows. Artex can continuously handle scanning, attempts, pivoting, and data collection without defenders having to confirm each step one by one. The traffic and data it generates can even be plentiful enough to trigger enterprise alerts, but by the time staff arrive on the scene, the attack may already be over.
Shutting down the old version does not eliminate the risk either. The developer switched the software to closed source after Artex was used in attacks, but the old open-source version is still circulating. Enterprises still have to deal with old credentials, exposed entry points, and weak protections — gaps that can still leave attackers a path.
Why Artex can run a full attack chain in three hours, how AI specifically chooses and executes the next step, and which credentials and exposed systems enterprises should address first are detailed below.
The Remaining ~5 Minutes Are for Members.
The free portion ends here. Behind the wall there is also:
Behind the wall: the alert has just lit up, and the AI hacker has already pried the door open.