FBI Warns Hackers Use Autonomous AI Tool Artex to Breach Networks in Under Three Hours
Technology news
The FBI warned that hackers are deploying Artex, an autonomous AI-driven penetration testing tool, to compromise enterprise networks in under three hours. By linking to frontier large language models, the software scans, exploits basic misconfigurations, and moves laterally before security teams can respond, impacting firms such as ConnectWise and Barracuda.
FBI Warns of Artex AI Tool Breaching Networks in Under Three Hours
The Federal Bureau of Investigation (FBI) has issued an advisory warning that malicious actors are utilizing an AI-driven penetration testing tool named Artex to automate network intrusions. The software independently handles network scanning, exploitation, and lateral movement by connecting to large language models (LLMs). According to the FBI, in some observed cases, the tool progressed from initially targeting a system to active compromise "in under three hours."
The tool has already been tied to several significant intrusions. Confirmed targets include IT management software provider ConnectWise—affecting the Microsoft 365 cloud environments of at least 2,250 customers—security vendor Barracuda, and several South Korean financial institutions where personal records of at least 68,000 individuals were stolen.
Autonomous Capabilities and LLM Integrations
Originally built as an open-source platform for defensive security teams to stress-test their networks, Artex has been repurposed outward by attackers. Operators configure Artex by connecting it to hosted frontier models from OpenAI and Anthropic, various open-source models, or models hosted outside the United States.
Once deployed, Artex typically begins by rapidly scanning target networks for basic vulnerabilities, such as exposed or stolen credentials and weak access protections, to secure initial access. It then autonomously pursues sensitive data stores, including databases and email accounts. Although the FBI characterized Artex operations as "noisy"—generating high volumes of traffic and defensive alerts—the rapid execution allows the tool to spread across networks faster than human security teams can typically investigate.
Targeted Sectors and Documented Compromises
The FBI reported that threat actors have deployed Artex against technology and defense firms, government agencies, financial services companies, internet and telecommunications providers, and legal firms.
In the Barracuda incident, financially motivated attackers used Artex to locate an unrevoked third-party service credential left active after a vendor relationship ended. This credential granted access to stored customer data-collection credentials. Barracuda noted that because its logs lacked sufficient detail to rule out unauthorized access, it rotated all customer credentials and directly advised a minority of its XDR customers to recreate integration credentials. In the ConnectWise incident, Microsoft attributed the attack to suspected Chinese actors seeking customer data.
Tool Limitations, Security Flaws, and Defensive Realities
Importantly, the FBI confirmed that none of the intrusions investigated thus far utilized new software exploits. Instead, Artex succeeded entirely by discovering standard security oversights: unrevoked vendor credentials, leaked passwords, and the absence of multi-factor authentication (MFA).
Although Artex's developer turned the repository into closed-source software following public reports of abuse, the FBI noted this change came too late, as earlier open-source iterations remain widely accessible. Furthermore, security researchers recently discovered hundreds of Artex instances exposed on the public internet without password protection due to configuration errors—including the instance used in the South Korean financial breaches. These exposed systems potentially allow outside actors to take control of ready-made offensive setups without building their own infrastructure.
Reported by techspot.com: https://www.techspot.com/news/114579-fbi-hackers-using-autonomous-ai-tool-break-networks.html