At Meta we’re committed to helping build an ecosystem of open-source tools for securely deploying large language models. In this presentation, we’ll introduce the tools we’ve open sourced so far, covering our approach to assessing and mitigating security risks in large language models. Specifically, we’ll cover our approach to assessing risks like prompt injection, code interpreter abuse, insecure code production, and model-facilitated cybercrime. We’ll end by describing how you can use our tools in your own LLM security work.