· via TechCrunch
FBI warns open-source AI hacking tool Artex breached networks in under three hours
The FBI says the autonomous AI tool Artex has moved from targeting to active compromise in under three hours, hitting ConnectWise customers, South Korean banks and security firm Barracuda.

FBI flags autonomous AI attacks
The FBI has issued a flash cybersecurity advisory warning that hackers are carrying out autonomous, AI-driven attacks against organizations in the United States and around the world, according to TechCrunch. The advisory puts a spotlight on Artex, an open-source, AI-powered penetration testing platform that attackers have wired to hosted frontier models from Anthropic and OpenAI, as well as to open-source and non-U.S. models.
According to the bureau, campaigns in recent months have struck financial services, government, technology, defense, telecom and law firm targets. In some incidents, Artex-linked intrusions moved from targeting a system to active compromise in under three hours.
How the tool operates
The FBI describes a simple but fast playbook: Artex rapidly scans a victim's network for signs of weakness, then exploits basic security problems such as weak protections and exposed or stolen passwords to gain a foothold. Once inside, it spreads through the network looking for databases, mailboxes and other sensitive information.
The bureau adds that Artex intrusions can be "noisy," generating large volumes of traffic and data that should trip a target's defensive alerts. The catch is speed: the tool lets attackers operate at a pace human responders struggle to match.
Confirmed victims so far
According to TechCrunch, Artex has appeared in several attacks since its public debut in July.
In South Korea, the tool was used against a number of financial institutions, with intrusions aided in large part by weak defenses, including missing multi-factor authentication. Those breaches exposed the personal data of at least 68,000 people.
The FBI also confirmed earlier Microsoft findings that attackers used Artex to break into IT management company ConnectWise and reach the Microsoft 365 cloud environments of at least 2,250 of its customers. Microsoft assesses the hackers as likely Chinese and says data was stolen from the affected customers. TechCrunch reports that ConnectWise did not respond to a request for comment.
Security firm Barracuda separately disclosed its own Artex-linked breach by financially motivated criminals. Per TechCrunch, the tool found a credential belonging to a third-party service provider that Barracuda had failed to revoke after ending the relationship. That access exposed stores of credentials Barracuda uses to gather data from customer systems. Because its logs were not detailed enough to rule out misuse, Barracuda said it chose to assume the worst and reset all customer credentials. Chief Information Security Officer Arve Kjoelen wrote that defenders who rely on spotting malicious activity and then responding "will often arrive after the attacker has finished." Barracuda did not say how many customers are affected.
The code is still out there
After Artex surfaced in real-world attacks, its developer made the software closed-source to prevent further abuse. The FBI says that ship has sailed: the earlier open-source version continues to circulate and remains, in the bureau's words, widely accessible to attackers.
The infrastructure around the tool is leaking too. Researchers last week found hundreds of publicly exposed Artex systems sitting on the internet without passwords, apparently through mistake or misconfiguration. One of them was the Artex system used in the South Korea attacks, meaning outsiders may be able to reach powerful hacking capabilities directly.
Why it matters
The FBI's advisory is one of the clearest signals yet that offensive security automation has become commodity tooling. Artex is not breaking exotic zero-days; it is finding weak passwords, missing MFA and forgotten vendor credentials faster than organizations fix them, and the victims so far — banks, an IT management giant and a security vendor — show that even well-resourced defenders are exposed. Two implications stand out. First, the speed asymmetry: detection-and-response security models assume humans can react in time, and an attacker that goes from scan to compromise in under three hours compresses that window dangerously. Second, the dual-use problem is now concrete — an open-source pentesting project that debuted in July was tied to major breaches within months, closing the repository did not recall the code, and the systems running it are themselves exposed. For defenders, the FBI's note about noisy intrusions is the actionable part: this class of tool generates detectable traffic, so aggressive monitoring for high-volume scanning, prompt credential revocation and enforced multi-factor authentication remain the practical countermeasures.
- #ai-security
- #cybersecurity
- #fbi
- #penetration-testing
- #open-source