Research paperAvailable internationally
ICLR 2024 paper introducing SWE-bench: 2,294 real GitHub issues from 12 Python projects as a test bed for language models.
Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations research paper summary (October 2026)
Meta paper on Llama Guard, a language model that classifies chatbot prompts and responses against a safety risk taxonomy.
What is Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations?
The Llama Guard paper from Meta presents a language model that acts as a safeguard for chatbots by classifying prompts and responses against a risk taxonomy. Newer Llama Guard versions and related tools are published in the Purple Llama repository.
The original Llama Guard is a Llama 2 7B model instruction-tuned on a curated dataset for prompt and response classification. The paper reports performance that matches or exceeds content moderation tools of the time on benchmarks such as the OpenAI Moderation Evaluation dataset and ToxicChat.
The Purple Llama repository now holds Llama Guard 3 in 8B, 1B and 11B vision variants, Prompt Guard against prompt attacks, Code Shield and the Cyber Security Eval benchmarks. Evals and Code Shield are under MIT, the models under Llama community licenses. The repository had 4,407 stars and was updated on August 18, 2026.
Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations AI features and use cases
- Filters chatbot inputs and outputsLlama Guard checks each user message and each model answer against defined risk categories and flags violations. Teams running internal or customer-facing assistants add a moderation layer they can host themselves.
- Blocks prompt injection attemptsPrompt Guard classifies inputs for prompt injection and jailbreak patterns. Applications that feed documents or emails into a model can catch hidden instructions before they reach it.
- Adapts categories to company policyBecause the taxonomy is part of the prompt, teams can adjust categories to their own content rules. Compliance and security teams keep moderation aligned with internal guidelines.
What sets Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations apart
- Open-weight safeguard models that can run on your own infrastructure.
- Small 1B variant for low-latency moderation.
- Model weights under Llama community licenses, tools and evals under MIT.
Where Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations is strongest
United States Available internationally
The paper is freely available on arXiv, and the accompanying code runs on your own hardware in any country.
Questions to ask before you buy Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations
Check the Llama community license terms for each model version before commercial use. A classifier makes mistakes in both directions, so measure false positives and misses on your own conversations and keep human review for escalations.
Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations reviews
No reviews yet. Be the first to share how your team uses it.
Every review shows its verification level. We read each one before it goes live. How we check reviews
Questions about Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations
Which Llama Guard versions exist?
The Purple Llama README lists Llama Guard, Llama Guard 2 and Llama Guard 3 in 8B, 1B and 11B vision variants, plus Prompt Guard.
Which licenses apply?
Evals and Code Shield are under MIT. Llama Guard models are under the corresponding Llama community licenses, as listed in the README.
Does Llama Guard work with non-Llama models?
Yes. It classifies text prompts and responses, so it can sit in front of or behind any chat model.
Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations update historyno updates yet
No changes recorded for Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations yet.
Changes the vendor made and confirmed user reports, each checked against the vendor's own pages. We review every tool monthly.
Alternatives to Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations
Tools in the same category (Safety & evaluation) with the most overlapping markets.