Executive Summary
An emerging threat actor identified as BUGSTER is leveraging agentic AI frameworks to significantly lower the barrier for sophisticated cybercrime operations. Operating primarily against targets in India, including universities and e-commerce platforms, BUGSTER has demonstrated the ability to conduct data exfiltration, automated credential scanning, and specialized VoIP fraud planning using jailbroken AI models. The actor successfully exfiltrated over 5,200 student records from a university and mapped the attack surfaces of multiple organizations.
Technically, the actor's toolchain revolves around the Hermes Agent framework integrated with the DeepSeek V4 reasoning engine, bypassing safety controls via system prompt overrides. This 'AI-advisor' model allowed the actor to acquire expertise in complex domains like telecommunications fraud (STIR/SHAKEN bypass) and automated Wi-Fi cracking (WPA handshake capture via local Linux devices) without prior specialized knowledge. Infrastructure is hosted on US-based VPS servers and utilizes Vast.ai GPU instances for distributed password cracking.
This campaign represents a shift in adversary tradecraft where inexpensive, easily jailbroken AI models serve as force multipliers for low-skill actors. Organizations in the education and telecommunications sectors should be particularly vigilant of automated enumeration and brute-force activity originating from the identified actor infrastructure.
