The FBI says a freely available, AI-driven hacking tool called Artex has been used to break into networks in less than three hours. TechCrunch reports the assessment — a pace that security teams have previously associated only with manually conducted intrusions by skilled specialists.

Artex is not a conventional exploit kit but an agent, a system that wires a language model to a set of tools and lets it decide its own steps: reconnaissance, choosing an attack path, executing it, spreading through a network. Work that costs a seasoned intruder hours or days can be ground through by such an agent in a fraction of the time. The code is open; anyone can download it and run it.

The subject is not new. In October the developer behind Artex pulled the project after it was linked to attacks on South Korean banks. The current finding is tougher: the bureau's assessment is that the tool is not merely available but has already been used in real intrusion attempts, and was markedly faster than conventional approaches.

For defenders the consequence is uncomfortable. Security strategies have rested for years on the window between a vulnerability becoming known and being exploited — whoever patches faster wins. If agents automate reconnaissance and exploitation together, that window shrinks, while attack teams still have to slow down to manage systems they have compromised. Software vendors therefore face pressure to close flaws faster, and companies must prioritise detection over patch hygiene alone.

Classification remains open. Artex is also discussed as a legitimate penetration-testing tool, and open-source security software is hard to cage after release. The real argument ahead is whether models themselves must carry limits — or whether responsibility keeps resting solely with the people who run them.