Zach Stein-Perlman comments on DeepMind: Evaluating Frontier Models for Dangerous Capabilities