AI‑driven self‑replicating worm shown by Toronto team

Researchers have demonstrated a new form of self-replicating malware powered by publicly available artificial intelligence, a development they warn could significantly lower the barrier for sophisticated, autonomous cyberattacks. The proof-of-concept computer worm, developed in a secure, isolated laboratory at the University of Toronto, can analyze its digital surroundings, tailor its attack strategy to different devices, and spread across a network without human intervention.

Unlike traditional worms that rely on a fixed set of instructions to exploit specific software flaws, this AI-driven prototype uses an open-weight large language model to reason about its targets. It was tested in a simulated network containing Linux and Windows machines as well as Internet of Things devices like printers and cameras. The worm successfully exploited known, unpatched vulnerabilities and common misconfigurations, gathering data such as passwords and system information to identify further attack paths as it spread.

A key feature of the worm is its ability to sustain itself by hijacking the computing resources, including graphics processing units, of infected machines to run its AI model. This...