Import AI 467: Self-sustaining AI viruses; pacing AI progress; confusion about AI and creativity

Researchers from the University of Toronto, Vector Institute, University of Cambridge, and ServiceNow report a proof-of-concept AI-driven computer worm that uses compromised GPUs to run an open-weight language model, identify vulnerabilities, tailor attacks, and replicate across hosts without relying on external APIs. It detected vulnerabilities about 80% of the time, exploited them about 53% of the time, and replicated about 88% of the time, producing an overall full-attack success rate of roughly 37% in the researchers’ evaluation. The result suggests self-sustaining AI-enabled cyberthreats are no longer purely theoretical, although the system remains imperfect and depended on a custom tool harness. Separately, about 1,337 employees and leaders from major AI companies signed a statement asking the U.S. government to support international technical and governance tools for deliberately pacing frontier AI development. They argue that potential automation of AI research could accelerate capability progress beyond society’s ability to understand or control it, while competitive pressures make unilateral slowdowns difficult. Early “shadow evaluations” of AI agents conducting two not

Follow Import AI to make it a durable For You signal.