AI-Powered Worm Leverages Stolen Compute to Target Linux, Windows, and IoT Devices

A research proof-of-concept demonstrates an AI-powered worm that runs a local LLM on compromised GPU hosts to autonomously scan, reason about, and exploit heterogeneous networks (Linux, Windows, IoT), parasitically stealing GPU compute to provide reasoning for constrained devices; in a 33-host testbed the agent identified ~82% of eventual targets, compromised ~75% of hosts on average, and achieved up to seven generations of self-replication while dynamically generating exploits from fresh advisories.