Artificial Intelligence Agent Attack For the first time, positive technology experts have investigated a cyber attack on a Russian company that was allegedly carried out entirely by artificial intelligence agents. The incident affected infrastructure on Linux and Windows. According to the PT ESC IR department, about three hours passed from penetration through an external service to the leakage of confidential data from internal systems. During this period, the attackers gained access to more than ten nodes of the corporate network. The head of the department, Denis Guidenko, noted that the actions of the AI ​​agents were fast and illogical from a human point of view: they tried different promotion methods at the same time, made mistakes and corrected them immediately. This makes the attack unusually fast and at the same time noticeable. Initial access was obtained through an error in the external service configuration, which is difficult to detect manually. Experts believe that AI agents can independently go through the entire attack chain - from penetration to data leakage. The company warned that the use of AI agents complicates attribution: tools are created and changed during an attack. This changes the requirements for protecting corporate infrastructure, and reduces the time defenders must detect and stop an attack.

Artificial Intelligence Agent Attack For the first time, positive technology experts have investigated a cyber attack on a Russian company that was allegedly carried out entirely by artificial intelligence agents. The incident affected infrastructure on Linux and Windows. According to the PT ESC IR department, about three hours passed from penetration through an external service to the leakage of confidential data from internal systems. During this period, the attackers gained access to more than ten nodes of the corporate network. The head of the department, Denis Guidenko, noted that the actions of the AI ​​agents were fast and illogical from a human point of view: they tried different promotion methods at the same time, made mistakes and corrected them immediately. This makes the attack unusually fast and at the same time noticeable. Initial access was obtained through an error in the external service configuration, which is difficult to detect manually. Experts believe that AI agents can independently go through the entire attack chain - from penetration to data leakage. The company warned that the use of AI agents complicates attribution: tools are created and changed during an attack. This changes the requirements for protecting corporate infrastructure, and reduces the time defenders must detect and stop an attack.