IBM becomes first major open-source AI model developer to earn ISO 42001 certification
IBM is among the first to earn this recognition, and the first open-source AI model developer to do so for the AI Management System (AIMS) of its IBM Granite open-source language models.
The International Organization for Standardization (ISO) introduced the world’s first standard for managing artificial intelligence in 2023. Known as ISO/IEC 42001:2023, it was designed to guide how organizations establish policies, processes, and safeguards for AI. For the first time, there is a globally recognized framework to externally audit how AI is responsibly built, deployed, and maintained.
IBM has achieved accredited certification under ISO/IEC 42001:2023 for the AI Management System (AIMS) of IBM Granite. The certification applies to IBM Granite language models; our family of open-source language models custom built for the enterprise.
From the beginning, Granite has been developed in the open and released under a permissive Apache 2.0 license. This open approach has been recognized externally with Granite earning a top five rank on Stanford’s Foundation Model Transparency Index, outperforming nearly all major model developers. This combination of openness, transparency and certified governance provides enterprises with greater confidence in adopting Granite for their most critical workloads.
“We’re honored to earn ISO 42001 certification for our flagship Granite models,” said David Cox, VP, AI Models. “Granite is now the first open model family to meet that bar, and it is a testament to the care that goes into building and maintaining them.”
The certification audit was conducted by Schellman, a leading accredited certification body. “Achieving ISO 42001 certification is a major milestone, not just for IBM Granite, but for the artificial intelligence and technology landscape,” said Avani Desai, CEO of Schellman. “As one of the first open-source AI model providers to be certified, IBM has now set an important precedent for how transparency, accountability, and innovation can coexist. Our entire team at Schellman is proud to be IBM’s partner on this journey and see firsthand their commitment to building AI responsibly.”
AI doesn’t currently behave like traditional software. A conventional program follows fixed instructions and produces the same deterministic output. Whereas AI models learn patterns from data and generate responses probabilistically. That means the output can vary, even when the input looks the same. This flexibility is powerful, but it also creates risks that must be managed throughout the entire model development and deployment lifecycle.
ISO 42001 establishes a framework for how organizations manage artificial intelligence. It defines what an AIMS should look like: policies, practices and safeguards that guide AI development and deployment. The standard is designed to address questions of ethics, accountability, and trust. These are questions that every organization adopting AI faces, particularly given increased regulatory scrutiny worldwide.
For enterprises, governments and developers, this certification provides confidence that IBM has implemented measurable safeguards and robust, transparent processes. It gives organizations building with Granite a trusted foundation for adopting the language models in high-stakes contexts, including highly regulated industries like financial services and healthcare, and the public sector, with many mission-critical deployment environments.
For AI model developers, the standard represents an expectation of what responsible AI looks like in practice. And for IBM, it validates years of investment in our AI security, safety, and governance practices, in addition to our industry-leading transparency around AI model disclosure.
More broadly, IBM’s achievement sets an important precedent that the future of responsible AI deployment is open, trusted and permissively licensed to support widespread model availability and innovation. By sharing models, benchmarks and tools with the broader community, IBM is contributing to an ecosystem where transparency and accountability are the industry norm.
With more than 30 million downloads, Granite is a widely adopted model family among open-source developers. At the enterprise level, organizations are deploying Granite in production for a variety of use cases such as retrieval-augmented generation, customer service and agentic workflows. ISO 42001 certification provides an additional layer of trust for anyone building with Granite.
As AI adoption accelerates worldwide, ISO 42001 certification provides a strong framework to balance innovation with safeguards.
Certification under ISO 42001 requires a comprehensive audit of how an organization manages its AI systems including governance policies, data practices, and technical safeguards. This audit was conducted by Schellman, an accredited certification body recognized for its thoroughness and quality. Schellman is the market leader in ISO 42001 and the first certification body recognized by ANAB in 2024.
Renowned for expertise grounded in practical experience, Schellman’s professionals deliver exceptional client service while maintaining steadfast independence. This balanced approach has enabled IBM to build a trusted, long-term relationship with Schellman, one that has helped IBM successfully achieve multiple compliance objectives, including this landmark ISO 42001 certification.
The process was completed in under three months and resulted in zero non-conformities, a significant achievement and testament to IBM’s mature AI governance practices. The certification validates that IBM Granite aligns with internationally recognized best practices for safe and responsible AI and that our AI management processes meet the highest levels of scrutiny.
Granite reflects IBM’s philosophy that transparency, safety, security and governance must be core principles in trusted AI systems. These principles contributed to IBM achieving ISO 42001 certification, which represents a broader commitment.
In addition to industry-leading transparency, IBM applies rigorous governance and testing throughout Granite’s lifecycle. IBM’s Data Management Framework Lakehouse oversees more than 2.7 petabytes of training data with strict metadata tracking and license controls, ensuring accountability at every stage. Beyond certification requirements, Granite models are stress-tested against real-world threats, digitally signed for authenticity and supported by open benchmarks for evaluating safety and bias—measures that strengthen trust not only in Granite, but across the broader AI ecosystem.
Building on these measures, Granite Guardian, IBM’s suite of open-source guardrail models, provides specialized safeguards that detect risks in prompts and responses across any AI model. By layering Granite Guardian onto Granite’s core protections, IBM helps enterprises ensure safe and responsible use of AI in production environments.
Together, these efforts demonstrate IBM’s philosophy: enterprise AI demands enterprise-grade trust. Becoming the first major open-source AI model developer to achieve ISO 42001 certification validates Granite’s foundation of transparency, security, and governance. This is a significant milestone for IBM, and we are committed to continuous innovation and investment to set the standard for responsible AI.