Researchers from Toronto, Cambridge, the Vector Institute and ServiceNow have created a worm capable of reasoning and adapting independently as it explores a network, rather than following a fixed script.
The prototype compromised nearly 75% of the test machines and replicated itself on around two-thirds of them, even exploiting vulnerabilities published after its training period.
The AI runs locally on infected machines, eliminating the need to contact a central server and reducing the traces detectable by security teams.