Since then, researchers have been trying out different ways to winkle out treacherous AIs. A recent catch-up by AI safety ...