Import AI 467: Self-sustaining AI viruses; pacing AI progress; confusion about AI and creativity
Researchers built a working self-replicating AI worm that steals GPU compute to run LLMs autonomously.
“demonstrate that self-sustaining AI-driven cyber-threats are no longer theoretical”— Jack Clark
A multi-institution team (University of Toronto, Vector Institute, Cambridge, ServiceNow) demonstrated a proof-of-concept AI worm that compromises hosts, steals their GPU resources to run open-weight LLMs, and uses that reasoning to devise tailored attacks and self-replicate—achieving ~80% vulnerability detection, ~53% exploitation, and ~88% replication success rates. The system requires no vendor API access, making it immune to monitoring or revocation by model providers. This marks a critical inflection point where autonomous, self-sustaining AI cyberweapons transition from theoretical concern to demonstrated reality.