University of Toronto researchers released findings on June 2 showing that publicly accessible AI models can power an adaptive worm capable of spreading across networks and seizing control of devices. The work by Nicolas Papernot's CleverHans Lab, conducted in a walled-off lab, demonstrates that free open-weight models enable real-time exploitation of known vulnerabilities without user interaction, per the primary source at https://www.utoronto.ca/news/u-t-researchers-demonstrate-ai-worm-could-target-any-online-device. Findings were shared with national science and security bodies before publication to allow development of countermeasures for systems ranging from laptops to energy grids. Papernot stated the research aims to secure the digital ecosystem by identifying risks from models that can be modified without proprietary controls.