Open the AI “black box” with IBM Granite. Build and deploy AI solutions with confidence.
Responsible AI matters
Enterprise AI demands enterprise-grade trust. With some models trained on pirated data or producing biased outputs, it’s easy to see why it matters. IBM® Granite® models are built with security, safety and governance at their core, giving you the confidence to build responsible AI.
Built in the open
Granite models are built from the ground up for secure, enterprise deployment.
True open source development
Unlike many "open" models that release only model weights, IBM Granite provides extensive transparency, including training data sources, methodologies and architectural decisions under a permissive Apache 2.0 license.
Model Openness Framework
IBM Granite qualifies as a Class III Open Model on the Linux® Foundation’s framework classifying the completeness and openness of models, an achievement not met by many model providers.
Industry-leading transparency
IBM has achieved the top ranking on the Stanford Foundation Model Transparency Index.
Enterprise grade trust
Introducing Granite Guardian 4.1
Deploy anytime, anywhere, for any AI-native workload. Building on the robust safety and security features natively built into Granite models, Granite Guardian adds a layer of specialized safeguards. The guardrail models detect risks in prompts and responses from any AI model, helping to ensure safe and responsible deployment.
Comprehensive risk detection
Detects and mitigates hallucinations, harmful content, bias, jailbreaking attempts and RAG quality and accuracy issues across prompts, responses and agentic workflows.
Industry-leading performance
Granite Guardian models hold six of the top 10 spots on the GuardBench leaderboard, measuring how well guardrail models can detect harmful and hallucinated content as well as attempts to jailbreak safety controls.
Open source models, tooling and assets
At IBM, we don’t just build responsibly with IBM Granite, we open source the models, tools and assets that we use to support responsible, open source AI innovation.
Granite Guardian detects and mitigates hallucinations, harmful content, bias, jailbreaking attempts and RAG quality and accuracy issues across prompts, responses and agentic workflows.
ARES is an open-source framework leveraged by the IBM Granite team that enables developers and security researchers to define targets, craft adversarial payloads and assess AI behavior under various threat models to detect failures in AI system responses.
Policy Tools is an open-source repository for defining, testing and enforcing policies in GenAI applications. Using a simple YAML format, teams can specify what their AI can and cannot say, generate synthetic data for testing and verify compliance at runtime.
The Granite 4.1 model and data metadata disclosures are made available in machine-readable JSON format for simple extraction and parsing to inform risk-based controls at runtime in generative applications.
The Mellea Skills Compiler is an open-source tool for hardening agent skills specified in natural language so that they are intrinsically more safe, secure and efficient.
The AI Atlas Nexus is a taxonomy and tooling to help identify, assess and mitigate AI-related risks and enterprise implications.
Trusted collaborators
We don’t secure AI in isolation. IBM works closely with with leading experts and organizations to continuously test, validate and strengthen the safety and security of Granite.
Resources
Technical documentation, model cards and implementation guides for Granite models.
Comprehensive resource on how to leverage Granite models in a responsible way for enterprise use.
The Baseline acts as a minimum set of requirements for AI projects relative to maturity and risk level. It is designed as a foundational guide for open source AI developers.
Explore the essential guardrails and frameworks needed to deploy AI models responsibly at scale.
IBM’s core principles for responsible AI development and approaches to building trustworthy systems.
AI governance toolkit for managing, monitoring and ensuring compliance of AI models across hybrid cloud environments.
Technical blog for the Granite 4.1 language models.