A team of researchers from the University of Toronto and cybersecurity firm CleverHans created an AI worm that spreads autonomously using a local LLM to reason about vulnerabilities. Tested on a simulated network of 33 hosts, it compromised 62% within a week. The system adapts its attack strategy based on the target environment, unlike traditional worms that follow a fixed sequence. The worm uses a tiered design where GPU-equipped devices provide reasoning for lightweight agents on low-power IoT devices.