MIT researchers created a new data privacy metric, Probably Approximately Correct (PAC) Privacy, and built an algorithm based on this metric that can automatically determine the minimal amount of randomness that needs to be added to a machine-learning model to protect sensitive data from an adversary.