“Sufficiently capable AI models will implicitly develop a theory of mind and an understanding of other agents, making it unnecessary to explicitly model them as distinct from the environment.”