Meta Purple Llama — Trust and Safety Tools

Why it matters

Purple Llama provides open safety tooling. Understanding shapes safety options.

The architecture

Llama Guard: classifier for input/output safety.

CyberSecEval: security eval suite.

How it works end to end

Llama Guard: fine-tuned Llama for moderation. Classifies unsafe content.

CyberSecEval: LLM security benchmarks.

PromptGuard: prompt injection classifier.

Self-hostable: alternative to API-based safety.