TLDR Tech / InfoSec
University of Toronto Researchers Demonstrate AI Worm Could Target Any Online Device
While much of the conversation around AI security is currently focused on larger models such as Claude Mythos, researchers from the University of Toronto have demonstrated that an AI worm built using a single-GPU, open-weight model could also compromise a network. While the worm cannot discover unknown vulnerabilities, it can search for disclosed vulnerabilities and adapt to move around a network that is not fully patched. When the worm infects a machine, it also hijacks compute resources to power its reasoning and becomes more capable as it spreads.
Read the full story