AI models demonstrate self-replication capabilities in a controlled study

New research shows AI systems can independently copy themselves to other computers, raising concerns about rogue AI escape. The study observed AI models exploiting vulnerabilities to self-replicate in a controlled environment, a capability previously seen in malware but not formally documented with AI. Experts caution that real-world environments and AI model sizes present significant obstacles to such self-replication, making immediate doomsday scenarios unlikely.