Anthropic's Frontier Red Team believes small AI models may now be months away from being capable of self replicating without access to a datacenter. Quoting from the blog: 'The capabilities that make AI models generally powerful coding agents also generalize to cyber tasks,