THE CRUNCH

The FBI has warned that hackers are using Artex, an open-source AI-powered penetration testing framework, to automate attacks on organisations worldwide. In a flash advisory, the bureau says some intrusions progress from initial targeting to privileged access in under three hours, a pace that per the FBI's framing leaves human defenders struggling to keep up. Amazon Threat Intelligence issued its own alert describing Artex-driven campaigns at what it calls industrial scale.

Artex is one of several open-source AI penetration testing platforms, software built to probe a network's defences. It can be connected to hosted frontier models from providers such as Anthropic and OpenAI, as well as open-source and non-US models. The FBI notes Artex intrusions can be noisy, generating traffic that may trigger defensive alerts, but says the tool lets attackers exploit weak protections and exposed or stolen passwords, then move through networks hunting databases, mailboxes and other sensitive data.

The advisory cites real damage. Attacks on South Korean financial institutions, blamed in large part on weak defences and missing protections such as multi-factor authentication, exposed the personal data of at least 68,000 people per TechCrunch, though the FBI alert cited by The Register put the figure at about 67,000. The FBI also confirmed Microsoft's earlier finding that hackers, likely Chinese according to Microsoft, used Artex to reach the Microsoft 365 environments of at least 2,250 ConnectWise customers. Security firm Barracuda disclosed its own Artex-linked breach, triggered by an exposed third-party credential it had not revoked; it could not rule out data access from its logs and changed all of its customer credentials.

Amazon Threat Intelligence adds a different angle. Between 3 October and 5 November, its MadPot honeypot sensors observed two Artex campaigns: one stealing AI API keys at scale to gain free model access, and another building real-time deepfake videos to trick identity verification checks at two cryptocurrency exchanges. In the deepfake attempts, the agents got within seven pixels of the target face position, but the vendor's server-side checks rejected every attempt.

Containment has proved difficult. Per The Register, the original developer removed the GitHub repository in October, and per TechCrunch the developer made the software closed-source to prevent further attacks. Neither step has contained the problem: the FBI says the earlier open-source version remains widely accessible, and internet scanning service Censys identified 1,465 hosts running an exposed Artex console as of 4 November, a figure the bureau calls likely significantly higher when including systems behind firewalls or in private networks.