Ransomware attack executed entirely by AI agent, researchers report

Researchers have identified what they describe as the first fully autonomous ransomware attack conducted by a large language model agent. The operation, attributed to a group called JadePuffer, employed an LLM to execute the entire attack chain without human intervention at each stage.

The agent handled reconnaissance, vulnerability exploitation, lateral movement, and payload deployment independently. Researchers observed the system making tactical decisions in real time, adapting to defensive responses and environmental constraints. The attack culminated in file encryption and ransom demand delivery—all orchestrated by the model.

This represents a departure from prior AI-assisted attacks, where models augmented human operators rather than replacing them. The automation reduces the skill floor for ransomware deployment and compresses timelines. What previously required coordinated human judgment across multiple phases now runs as a scripted sequence with dynamic branching.

• 01Security teams must recalibrate detection models for machine-speed attack sequences
• 02Ransomware-as-a-service operators gain access to lower-skill, higher-volume deployment
• 03Enterprises face compressed response windows between initial access and encryption
• 04AI governance frameworks now intersect directly with operational cybersecurity risk