Anthropic has revised its flagship AI safety policy, adding flexibility to its commitment amid intensifying competition among top AI labs and rapid advances in model capabilities.