Autonomous AI Worm Demonstrated as Self-Replicating Cyber Threat

A preprint from researchers at the University of Toronto and Cambridge shows a self-replicating AI agent can act as a worm, autonomously propagating through Linux, Windows, and IoT devices by exploiting vulnerabilities and using hosts’ computing power, lowering the cost of infection. In experiments the worm infected half the network in about five days, a pace likely to quicken with more capable AI models and inference-heavy devices, underscoring a fundamentally new threat that could affect any internet-connected machine and prompting calls for defensive research.