Create and Implement Data Secure AI Workflows

Introduction

The novel, unpredictable nature of Large Language Model (LLM) applications can raise a lot of questions around what is happening to your data when you use them. Consider whether the data is private, where it might unexpectedly appear later, and how to ensure new leaks do not appear as you add agentic capabilities to our apps. This is especially important for apps that require legal compliance such as HIPAA, COPPA, GDPR, SOC 2, and more.

If built correctly, LLM applications have the potential to be as secure or more secure than conventional applications. LLMs can often be contained on a single server or run locally, depending on model size and hardware availability. Many tasks that may have required external tools in the past can now be completed with a call to a single model. However, with new methods of using data come new vulnerabilities that need to be considered. This tutorial discusses key considerations and strategies for building LLM workflows with data security in mind.