The European Union calls for robot rights but still wants a kill switch
Gizmag has a much more detailed page:
here is an excerpt:
The idea of mandating manufacturers implement a form of "kill switch" into their designs is not new. In 2016 researchers at Google DeepMind proposed what they called a "big red button" that would prevent an AI from embarking on, or continuing, a harmful sequence of actions. The paper Google released discussed the problems with implementing such a form of kill switch into a machine with self-learning capabilities. After all, the AI may learn to recognize the actions that its human controller is trying to subvert and either avoid undertaking similar tasks causing it to become dysfunctional or, in a worst-case scenario, learn to disable its own "big red button."
The Google DeepMind researchers suggested that any robot programmed with a kill switch would also need to be programmed with a form of selective amnesia that causes it to forget that it had ever been interrupted or usurped. This would stop the robot gaining awareness of its lack of autonomy.
EU to debate robot legal rights, mandatory "kill switches" emphasis added
Note that both these articles were dated 01/13/2017