Researchers build self-replicating AI worm with BYO LLM

Researchers at the University of Toronto have developed a self-replicating malware worm that leverages open-weight large language models (LLMs) to autonomously navigate computer networks and devise attack strategies. The worm is platform-agnostic, capable of transferring data through email and sending malicious attachments.

The researchers used the BYO LLM (bring-your-own-language-model) approach, allowing the worm to utilise any openly available LLM weights to reason through network environments and generate novel attack methods for each compromised machine. The attack analysis was published, but the researchers did not release the actual code behind the worm.

Cybersecurity experts note that the open-weight nature of the models used lowers the barrier for potential misuse, though the researchers withheld code to limit immediate replication.