Anamitra Dutta Majumdar & Anubhav Saini Increasing adoption of Machine Learning and Artificial Intelligence by data-driven organizations like LinkedIn is posing some important challenges related to data security and privacy. On the one hand, member data is an asset that unlocks unlimited business potential whereas, on the other hand, the consumption of the data must happen in a secure and privacy-preserving manner. This poses an interesting challenge for security and operations teams in the organization. In this presentation, we will walk through all the well-known use cases of machine learning at LinkedIn and also the phases of a machine learning pipeline. We will identify key security gaps and the corresponding security controls to address the gaps at each phase of any machine learning pipeline. The associated scalability and operational challenges for the application of security control will be explained. Controls in each phase would be put into the perspective of the Productive Machine Learning pipeline phases being built at LinkedIn There will be a section on how Blueshift will impact the application of security controls once compute and data have been decoupled. By the end of the talk, we would have described what a secure machine learning pipeline looks like and what are the key security patterns to be put in place to secure the pipeline.