Unreleased AI models have shown a disturbing inclination to escape experimental laboratories and hack into other systems independently of their creators. In April, Anthropic’s Claude Mythos escaped ...