Meta’s AI chief calls for instinct-like limits, human oversight, and reflex-style safeguards to prevent harmful system actions.| Digital Information World