> ## Documentation Index
> Fetch the complete documentation index at: https://docs.promptguard.co/llms.txt
> Use this file to discover all available pages before exploring further.

# Third-Party Models

> The open models PromptGuard detection runs on, their licences, and the terms that pass through to you

PromptGuard's detection engine calls a small number of openly licensed machine
learning models. This page names every one of them, says which licence it is
under, and carries the attribution and use terms those licences require us to
give you.

<Info>
  **Built with Llama.** PromptGuard's prompt-injection detection uses Meta's
  Llama Prompt Guard 2.
</Info>

## How we use these models

The engine sends text to each model through an inference endpoint; no
PromptGuard release ships model weights today. If a release starts to, this page
will say how each licence's text travels with the weights before that release
ships.

## Detection ensemble

These run on scans as part of the toxicity and prompt-injection ensemble.

| Model                                           | Used for                                          | Licence                          |
| ----------------------------------------------- | ------------------------------------------------- | -------------------------------- |
| `meta-llama/Llama-Prompt-Guard-2-86M`           | Prompt-injection and jailbreak screening          | Llama 4 Community License        |
| `protectai/deberta-v3-base-prompt-injection-v2` | Prompt-injection screening                        | Apache License 2.0               |
| `KoalaAI/Text-Moderation`                       | Content moderation, including hate and harassment | OpenRAIL                         |
| `unitary/toxic-bert`                            | Toxicity and identity-based attacks               | Apache License 2.0               |
| `cardiffnlp/twitter-roberta-base-hate-latest`   | Hate speech                                       | Creative Commons Attribution 4.0 |

## Other models

| Model                                    | Used for                                                 | Licence            |
| ---------------------------------------- | -------------------------------------------------------- | ------------------ |
| `openai/gpt-oss-safeguard-20b`           | Content-safety and multi-turn judging of escalated scans | Apache License 2.0 |
| `openai/gpt-oss-20b`                     | Jailbreak, RAG-poisoning and tool-injection judging      | Apache License 2.0 |
| `sentence-transformers/all-MiniLM-L6-v2` | Embeddings for multi-turn drift and response caching     | Apache License 2.0 |
| `mistralai/Mistral-7B-Instruct-v0.2`     | Turning natural-language policy descriptions into rules  | Apache License 2.0 |

## Licence terms

### Llama 4 Community License

Llama Prompt Guard 2 is made available under the Llama 4 Community License.
Llama 4 is licensed under the Llama 4 Community License, Copyright © Meta
Platforms, Inc. All Rights Reserved.

* The licence: [https://www.llama.com/llama4/license/](https://www.llama.com/llama4/license/)
* Your use of PromptGuard must also comply with Meta's Llama 4 Acceptable Use
  Policy, which the licence incorporates and which we pass through to you:
  [https://www.llama.com/llama4/use-policy/](https://www.llama.com/llama4/use-policy/)

### OpenRAIL

`KoalaAI/Text-Moderation` is released under an Open Responsible AI License
(OpenRAIL); its model card declares the licence family without naming a specific
OpenRAIL variant. OpenRAIL licences permit commercial use but attach
use-based restrictions that bind everyone downstream of the model, including
you. You must not use PromptGuard, or the moderation results it produces, for
any use those restrictions prohibit. The OpenRAIL licences and their
restrictions are published by the RAIL initiative:
[https://www.licenses.ai/ai-licenses](https://www.licenses.ai/ai-licenses)

### Creative Commons Attribution 4.0

`cardiffnlp/twitter-roberta-base-hate-latest` is by Cardiff NLP (Cardiff
University), published at
[https://huggingface.co/cardiffnlp/twitter-roberta-base-hate-latest](https://huggingface.co/cardiffnlp/twitter-roberta-base-hate-latest)
under the Creative Commons Attribution 4.0 International licence:
[https://creativecommons.org/licenses/by/4.0/](https://creativecommons.org/licenses/by/4.0/).
PromptGuard uses the model unmodified.

### Apache License 2.0

The models listed under the Apache License 2.0 are used under its terms:
[https://www.apache.org/licenses/LICENSE-2.0](https://www.apache.org/licenses/LICENSE-2.0).
None of their model repositories publishes a NOTICE file.

## Models we stopped using

`facebook/roberta-hate-speech-dynabench-r4-target` was removed from the
detection ensemble in September 2026 because it is published without a licence.
Its place is held by `cardiffnlp/twitter-roberta-base-hate-latest`.