Google's AI team is developing 'big red button' to switch off threatening systems

Google's DeepMind teamsay AI agents are 'unlikely to behave optimally all the time' and have called for 'safe interruptibility' to be built into systems.