The pervasiveness of the digital society has created a wealth of useful but potentially sensitive data. Artificial intelligence (AI) technology has become ubiquitous thanks to the availability of a solid digital infrastructure, a constant stream of data, and effective pattern-matching methods capable of making useful inferences on it. However, as an increasing number of AI agents is deployed for a wide range of real-world applications, concerns on their impact on privacy grow. Moreover, as these agents rely heavily on big data to perform their tasks, the need to protect personal information it contains is paramount. Regulation and their own business needs have prompted organizations to find new methods that address these concerns, leading to an interplay between privacy and artificial intelligence research. Privacy researchers have started to investigate the relationship between these two fields from two opposite vantage points. On the one hand, designers of privacy technologies have turned to AI technology to enhance the evaluation of their systems. On the other hand, privacy engineers have identified the need for AI agents to respect privacy, proposing their own methods to fill this gap. This thesis contributes to both strands of privacy research, focusing on producing practical methods that address specific limitations in the real-world deployment of AI technologies for privacy. On the one hand, we use novel AI technologies to improve the real-world evaluation of anonymous communications systems. On the other hand, we contribute new technologies and methodologies to engineer AI agents that respect privacy.