Compliance has become a key requirement for robots meant to interact with humans. It is viewed as a necessary property to increase safety and efficiency in human-robot cooperative actions. In humans, compliance takes three dimensions: mechanical, cognitive and social. While robotics has focused primarily on modeling the first two, we here discuss the importance to consider also the social dimension that compliance takes in human-human interactions and how this can be extended to human-robot interactions. We discuss situations in Which requesting that the human complies to the machine may be advantageous, and not the converse. We conclude with a list of open ethical and legal issues that may arise from developing actively non-compliant machines. (C) 2016 Elsevier B.V. All rights reserved.