UK AI Security Institute evaluations reveal that Anthropic's heavily restricted Mythos Preview model is not uniquely dangerous, performing…