More than 800 exposed instances of Artex, an open-source autonomous penetration testing tool developed in China, have been detected publicly accessible on the internet without basic authentication. Global cybersecurity firm Censys identified the unprotected installations, which allow direct access to active attack consoles through standard web browsers without requiring passwords. Silas Cutler, a principal security researcher at Censys, pointed out that although some instances could belong to authorized security testers, the software does not inherently require external internet access, leaving the reasons for widespread exposure unclear. Released in July 2026, the tool employs a multi-agent architecture designed to autonomously conduct reconnaissance, vulnerability scanning, system exploitation, and data exfiltration with minimal human supervision while bypassing safety filters found in closed frontier artificial intelligence models.
According to analysis from Microsoft, the automated multi-agent framework significantly outpaces conventional manual penetration testing methods. While human-driven tests often take days or weeks from reconnaissance to cloud compromise and operate at modest login speeds, Microsoft reported that Artex can achieve full cloud administrative control within three hours, executing automated login attempts at speeds reaching up to 2,900 requests per minute. Security researchers have traced the use of Artex to active cyberattacks across more than 200 organizations in 14 countries, including an intrusion into the American IT management platform ConnectWise.
The widespread availability of high-speed autonomous attack infrastructure highlights mounting operational challenges for corporate defenders worldwide. In South Korea, the framework was deployed in attacks against seven financial institutions between September and October 2026, resulting in the data breach of 67,000 users and 2,200 organizations. South Korean security firms AhnLab SEcurity intelligence Center and S2W TALON subsequently issued urgent countermeasures addressing the tool automated cloud and web server targeting mechanisms. As exposed control servers remain susceptible to third-party hijacking, enterprise defenses face increasing pressure to transition from manual monitoring and rule-based firewalls toward real-time automated mitigation capable of countering multi-agent intrusions.