Researchers Build Self-Replicating AI Worm That Operates Entirely on Local, Open-Weight Models

Security researchers at the University of Toronto have demonstrated a concerning evolution in autonomous cyber threats by developing a self-replicating artificial intelligence worm that operates independently within local environments. This proof-of-concept development signals a significant shift in the threat landscape, as it represents one of the first examples of an AI-driven malware capable of reasoning its way through networks without human intervention or reliance on commercial AI services.

The research team constructed this novel malware to operate entirely on locally hosted open-weight large language models (LLMs), allowing it to bypass the security controls typically implemented by commercial AI providers. The worm demonstrates an unprecedented capability to navigate network systems independently, analyze each target it encounters, and generate tailored attack strategies specific to the environment. Perhaps most alarmingly, the malicious code can self-replicate and propagate across systems, creating a potentially exponential threat that could rapidly spread through interconnected networks.

Organizations deploying open-weight AI models in their infrastructure face the most immediate risk from this development. While the current implementation remains a proof-of-concept, the techniques demonstrated could be adapted by malicious actors to create sophisticated, self-propagating malware that leverages the very AI systems intended to enhance business operations. The threat is particularly concerning because