Freedom251 › AI Content Moderation Software

Granite Guardian

Free plan APILinuxSelf-hosted

What Granite Guardian is

Granite Guardian is a set of AI safeguards for checking risks in prompts, responses, and agent workflows. It can detect hallucinations, harmful content, bias, jailbreaking attempts, and issues with retrieval-augmented generation (RAG) quality and accuracy. For RAG, it evaluates whether retrieved context is relevant, whether responses are grounded in that context, and whether answers are relevant. It can also examine function calls for syntactic or semantic hallucinations, including invalid calls and fabricated information.

Granite Guardian 4.1 offers a thinking mode with reasoning traces and a non-thinking mode that returns a judgment score. The 4.1 8B model supports user-defined judging criteria, such as formatting requirements, length constraints, and domain-specific instructions. IBM describes the models as suitable for AI-native workloads, with deployment through a vLLM container and options including watsonx.ai, Azure AI Foundry, Ollama, Replicate, LM Studio, Docker Hub, Dell, and AWS. The models use the Apache 2.0 license for research and commercial use. IBM says they are trained and tested only on English data and should be used in their prescribed scoring mode; reasoning traces may contain unsafe content and may not be faithful.

Who it suits

Granite Guardian suits teams adding risk checks to AI prompts, responses, RAG applications, or agent workflows. It may also fit organizations that need custom judging criteria or want to run models through options such as vLLM, Ollama, or LM Studio.

What it does well

Checks prompts, responses, and agent workflows for several risk types.
Evaluates context relevance, groundedness, and answer relevance for RAG.
Assesses agent function calls for invalid or fabricated information.
4.1 8B supports custom formatting, length, and domain criteria.
Apache 2.0 license allows research and commercial use.

What to know first

Models are trained and tested only on English data.
IBM advises using the models only in prescribed scoring mode.
Reasoning traces may contain unsafe content and may not be faithful.

Limits

The models are trained and tested only on English data. IBM says to use them strictly in the prescribed scoring mode; reasoning traces may contain unsafe content and may not be faithful.

Getting started

Visit the IBM Granite trust website.
Choose a deployment option such as vLLM, watsonx.ai, Ollama, or LM Studio.
For a local vLLM deployment, use the OpenAI-compatible chat completions endpoint.
Select the prescribed scoring mode for the intended checks.

Verdict

Choose Granite Guardian if you need AI risk scoring across prompts, responses, RAG, or agent function calls, with options for custom criteria. Look elsewhere if your workload requires non-English training coverage or use outside the prescribed scoring mode.

Questions

Is Granite Guardian free?

The listed pricing model is paid, and a free plan is also listed. The models are released under Apache 2.0 for research and commercial use.

What risks can Granite Guardian detect?

It checks for hallucinations, harmful content, bias, jailbreaking attempts, and RAG quality and accuracy issues across prompts, responses, and agentic workflows.

Can it evaluate RAG applications?

Yes. It assesses context relevance, response groundedness, and answer relevance.

What deployment options are available?

IBM lists watsonx.ai, Azure AI Foundry, Ollama, Replicate, LM Studio, Docker Hub, Dell, and AWS. IBM also provides instructions for running it in a vLLM container.

What license does Granite Guardian use?

The models are released under Apache 2.0 for research and commercial use.

Does Granite Guardian support custom policies or criteria?

Custom policies are listed, and Granite Guardian 4.1 8B supports user-defined judging criteria such as formatting rules, length constraints, and domain-specific instructions.

Facts from Granite Guardian's own pages

Agent workflows: It can assess function calls for syntactic and semantic hallucinations, including whether calls are valid and whether information is fabricated. · ibm.com, 9 Oct 2026
API: The IBM model selector shows a local vLLM deployment accepting requests at an OpenAI-compatible chat completions endpoint. · ibm.com, 9 Oct 2026
Custom criteria: Granite Guardian 4.1 8B supports user-defined judging criteria, including formatting rules, length constraints, and domain-specific instructions. · ibm.com, 9 Oct 2026
Deployment: IBM describes Granite Guardian as deployable for any AI-native workload and provides instructions for running it in a vLLM container. · ibm.com, 9 Oct 2026
Deployment options: IBM lists watsonx.ai, Azure AI Foundry, Ollama, Replicate, LM Studio, Docker Hub, Dell, and AWS as providers or options for running Granite models. · ibm.com, 9 Oct 2026
Founded: 1911 · ibm.com, 28 Sept 2026
Governance: IBM says its Granite AI Management System is ISO 42001 certified. · ibm.com, 9 Oct 2026
Headquarters: Armonk, New York, United States · ibm.com, 28 Sept 2026
Hosted option: The IBM model selector lists watsonx.ai, Ollama, and LM Studio as other places the model is available. · ibm.com, 9 Oct 2026
Intended use and limitation: IBM says the main Guardian models are intended for moderate cost, latency, and throughput use cases, while smaller models can suit stricter cost, latency, or throughput requirements. · ibm.com, 9 Oct 2026
License: The models are released under Apache 2.0 for research and commercial use. · ibm.com, 9 Oct 2026
Local platforms: IBM's partner page describes LM Studio as running models locally on Mac, Windows, and Linux, and Ollama as supporting devices including CPUs, GPUs, Apple M-series MacBooks, and Raspberry Pi. · ibm.com, 9 Oct 2026
Model limitation: IBM documentation says Granite Guardian is trained and tested only on English data. · ibm.com, 9 Oct 2026
Model security: IBM says Granite language models have cryptographic signatures to verify model provenance and undergo single-shot and adaptive multi-turn red teaming using ARES. · ibm.com, 9 Oct 2026
Operating modes: Granite Guardian 4.1 supports a thinking mode with reasoning traces and a non-thinking mode that outputs a judgment score. · ibm.com, 9 Oct 2026
Purpose: Granite Guardian adds specialized safeguards that detect risks in prompts and responses from any AI model. · ibm.com, 9 Oct 2026
RAG checks: For RAG use cases, it evaluates context relevance, response groundedness, and answer relevance. · ibm.com, 9 Oct 2026
RAG evaluation: For RAG use cases, Guardian assesses context relevance, response groundedness, and answer relevance. · ibm.com, 9 Oct 2026
Risk detection: It detects hallucinations, harmful content, bias, jailbreaking attempts, and RAG quality and accuracy issues across prompts, responses, and agentic workflows. · ibm.com, 9 Oct 2026
Security: IBM says Granite language models use cryptographic signatures to verify model provenance and undergo single-shot and adaptive multi-turn red teaming. · ibm.com, 9 Oct 2026
Usage caution: IBM says Guardian models should be used strictly in the prescribed scoring mode and that reasoning traces may contain unsafe content and may not be faithful. · ibm.com, 9 Oct 2026
Use limitation: IBM says the models should be used strictly for the prescribed scoring mode, and that deviation may lead to unexpected, potentially unsafe, or harmful outputs. · ibm.com, 9 Oct 2026

Where Granite Guardian stands on Freedom251

AI Content Moderation Software#7 of 27