University of Toronto uses free LLM to build spreading computer worm

Researchers at University of Toronto built a self-propagating computer worm using an unnamed open-weight LLM released in 2025. The worm adapts in real time to detect known vulnerabilities and exploits misconfigurations to move laterally across networks. It runs on a single GPU and operates without commercial AI safety features. The work shows attackers can now deploy AI-driven worms at low cost using publicly available models.