Once a prompt leaves your notebook and meets real users, new risks appear. People may try to trick the model, documents it reads may contain hidden instructions, and a wrong or unsafe answer can reach a customer. Safety and security work is about expecting these problems and limiting the damage.
You'll start with prompt injection, where untrusted input changes what the model does, both directly and through documents or web pages. Then you'll learn about guardrails, the checks placed around a model's inputs and outputs, and what it takes to run an LLM application in production.
That completes the course. Next, put what you've learned to work in the Projects section.
An introduction to direct and indirect prompt injection, one of the most common LLM vulnerabilities, and how to prevent it.
Why prompt injection is ranked as the top risk for LLM applications, from direct attacks to payloads hidden in documents.
What guardrails are and how these tools, processes, and guidance help keep generative AI models safe to use.
How to add guardrails when deploying large language models in enterprise environments.