Purple Llama
Meta's open toolkit of trust and safety tools for generative AI developers.
About
Purple Llama is Meta's umbrella project for open trust and safety tooling in generative AI. It gives developers the evaluations and safeguards needed to deploy models responsibly, starting with cybersecurity evaluations and input/output filtering. The project takes its name from purple teaming in cybersecurity, combining offensive and defensive approaches to AI risk.
Its components include CyberSecEval benchmarks for measuring LLM security, Llama Guard safety classifiers for content filtering, Code Shield for detecting insecure code generation, and LlamaFirewall for multi-layer defense. The tools are open source and developed with partners across the AI Alliance.
What it can do
- CyberSecEval LLM security benchmarks
- Llama Guard input/output safety classifier
- Code Shield insecure-code detection
- LlamaFirewall multi-layer defense
- Open source under MIT and Llama licenses
Setup
Clone the PurpleLlama repo from GitHub; components install via pip.
Prerequisites
Python environment; GPU optional for running guard models locally.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 126; Trending score 82/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.