A new analysis on LessWrong argues that capabilities research inherently expands the safety-usefulness Pareto frontier. Performance optimizations for inference allow developers to deploy weaker, safer models more extensively. This logic suggests that any research creating more options technically functions as safety research.