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

Researchers have built a prototype computer virus that uses open-weight artificial intelligence models to compromise computers and steal their GPU resources. The worm runs local reasoning graphs to detect vulnerabilities, exploit hosts, and autonomously self-replicate across networks without relying on vendor application programming interfaces. With an overall attack success rate of about thirty-seven percent, the worm creates a decentralized swarm of independent replicas that operate beyond single points of control.

In economic developments, experts suggest that smarter AI systems will drive up compute prices as they better monetize their work. At the same time, senior representatives from major Western AI labs have asked the United States government to support an international effort to build governance tools that can deliberately pace the frontier of automated AI development, helping society manage rapid capability acceleration and collective action problems.

Regarding research capabilities, new shadow evaluation studies show that current frontier AI agents struggle with the tasteful creativity needed for original academic research, often getting stuck on narrow paths or failing to incorporate feedback. However, OpenAI used an internal model to solve ten open problems in mathematics and computer science across high-dimensional geometry, quantum complexity, and group theory, demonstrating that AI can drive forward technical frontiers even as questions about independent creativity remain.