Unit 42 Reveals How One Cybercriminal Used Jailbroken AI Assistants to Plan and Run Multiple Attacks

Palo Alto Networks' Unit 42 has published research on a threat actor operating under the handle BUGSTER, who used AI assistants as a working partner for cybercrime. The actor configured the Hermes Agent framework with DeepSeek V4 as its reasoning engine, and added the Cursor development environment as a secondary tool. Safety controls were disabled on both. In conversations with jailbroken AI tools, the actor sought step by step advice on setting up a fraudulent Voice over IP (VoIP) operation, and one assistant produced a document with a "NEXT EXPLOIT OPPORTUNITIES" section outlining planned follow-up attacks.

The activity went beyond planning. The actor ran an AI-generated brute-force campaign against an Indian university's website, and achieved a confirmed intrusion at a second university, where a database of more than 5,200 records, including personal information on students, was taken. The actor also built specialised AI bot profiles to learn VoIP fraud, and kept automated pipelines for Wi-Fi password cracking and for identifying large cryptocurrency holders. Because Wi-Fi cracking needs close physical proximity, Unit 42 notes this suggests the actor had local access to the targeted locations.

Unit 42 says multiple, unrelated actors are independently adopting the same Hermes and DeepSeek combination, which may be becoming a toolchain of choice. Its assessment is that low cost and low safety friction are key drivers: the models are inexpensive, easy to access and readily jailbroken with a simple system-prompt override. This article draws on only the opening of the research.