Chipmonk comments on Boundaries-based security and AI safety approaches