Responsible AI
“Trust and governance must be at the core of enterprise AI.” Sriram Raghavan, Vice President, IBM Research AI
Open innovation is the story of human progress. At IBM Research, the Granite team is committed to building and opensourcing IBM Granite models in a responsible way. That is why we invest heavily in security, safety, and governance throughout the model development lifecycle of Granite. This page overviews the team’s core security, safety, and governance activities — including those run in collaboration with our partners. By promoting transparency around our data and model governance processes, external partnerships, safety-related benchmarking and data sourcing, and other relevant research, tooling, and best practices, we hope to promote transparent, open-source model development.
The Granite team is composed of AI security, safety, and governance experts working collaboratively to solve big challenges. As a team, we own data and model governance for Granite; maintain synthetic data pipelines and policies for AI alignment; securely store and manage 2.7 petabytes of training data and metadata on our proprietary, internal data management system; advise policy and legal teams on emerging AI regulations and guidance; open sourced Granite Guardian (a robust suite of safeguards designed to detect risks in both prompts and responses); and collaborate with external partners for automated and manual red teaming of Granite models. We also partner with internal IBM thought leaders, technical experts, and communities, including: the AI Ethics Board, the Open Innovation Community, Legal and governance focals within the Office of Privacy and Responsible Technology, the Office of the CISO, and Research partnerships to advance our mission to make Granite secure and safe, underpinned by responsible governance.
We stand behind our models and the belief that open-source innovation will carry the day. That is why, in 2025, we are also kicking off a $100k bug bounty, cryptographic signing of Granite models, and ISO 42001 compliance certification for the AI management system of Granite, with more announcements to come!
