Import AI 467: Self-sustaining AI viruses; pacing AI progress; confusion about AI and creativity
Researchers from institutions such as the University of Toronto and Vector Institute have successfully developed the first self-sustaining AI computer virus prototype. This virus utilizes the GPU resources of infected hosts to run open-source large language models, independently reasoning to detect vulnerabilities and customize attack strategies, thereby achieving self-replication. Its core mechanism consists of a reasoning graph composed of nodes for planning, evaluation, and action, breaking down cognitive process limitations to prevent context expansion. Tests show that the agent has success rates of approximately 80%, 53%, and 88% in vulnerability detection, exploitation, and self-replication, respectively, with an overall complete attack success rate of about 37%. The study proves that autonomous generative cyberthreats are no longer theoretical; the future Internet may evolve into a complex ecosystem composed of attacking and defensive AI agents.