This newsletter covers four distinct developments in AI research and policy. First, researchers from the University of Toronto, Vector Institute, Cambridge, and ServiceNow demonstrated a self-sustaining computer worm that uses open-weight LLMs on compromised GPUs to autonomously detect vulnerabilities and replicate.
Second, Dwarkesh Patel argues that as AI systems become smarter, compute prices will rise significantly, potentially exceeding $250k per H100 annually. Third, leaders from major labs including OpenAI, Anthropic, and Google DeepMind requested US government support for international tools to pace frontier AI development.
Finally, new research by Princeton, Cornflower Labs, and others suggests current AI systems excel at engineering but lack the creative intuition needed for autonomous scientific discovery.