Toronto researchers showed a small open-weight model could autonomously find, exploit, and spread across a 33-host test network, even adapting to newly disclosed flaws at runtime.

A new worm can autonomously spread through networks, raising alarms for cybersecurity teams and executives.

The demo makes one thing painfully clear: the same AI tools companies want for productivity can also help attackers scale exploitation of known software flaws.

A simple exploit of Meta's AI-powered support chatbot compromised the most guarded Instagram account in the world, exposing a systemic flaw in automated security.