Llama Guard

 Llama Guard is an open-weight safeguard model from Meta, released as part of the Llama protections suite. It classifies prompts and model responses against a taxonomy of safety categories, returning whether content is safe and which policy it violates, so it acts as a content filter around a primary model. Because the policy is supplied in the prompt, teams adapt the categories to their own application without retraining. Released with open weights, it runs privately as a filter around any generative model a team already operates. Multimodal variants extend the same screening to image content, and related models cover prompt and code safety.

Key Capabilities:

• Input and output classification against a safety taxonomy

• Customizable policy categories defined at inference time

• Multimodal variants that screen image content

• Open weights for self-hosted, private deployment

• Integration as a filter in front of any generative model

• Part of a suite alongside Prompt Guard and Code Shield