Researchers demonstrate AI-powered malware worm that adapts attacks across computer hosts

Summary

Researchers from the University of Toronto, Vector Institute, University of Cambridge, and ServiceNow have developed a proof-of-concept AI-driven worm that uses open-weight language models to autonomously adapt attacks across computer hosts. Unlike traditional malware that follows fixed exploit lists, this prototype tailors attacks to each target system. In seven-day autonomous lab runs, the worm averaged 31.3 vulnerabilities found and 23.1 hosts exploited. The authors warn that self-sustaining AI-driven cyber threats are no longer theoretical, highlighting how locally deployed open-weight models complicate malware containment efforts.