An article in IEEE Spectrum asks a question that hits close to home: robot safety functions already work, but what happens if the input lies? Perfect logic applied to a false world stays perfect, and wrong. I'm writing about it because it's my everyday trade.