At this time, we’re introducing the Databricks AI Governance Framework (DAGF v1.0), a structured and sensible method to governing AI adoption throughout the enterprise.
As organizations embrace AI at scale, the necessity for formal governance grows. Enterprises should align AI improvement with enterprise targets, meet authorized obligations, and account for moral dangers. This framework is designed to help program improvement, deployment, and steady enchancment.
The DAGF enhances the Databricks AI Safety Framework, providing a whole view of governance that spans each safety and operational integrity.
Why AI governance can’t wait
In keeping with a 2024 world survey of 1,100 know-how executives and engineers performed by Economist Impression, 40% of respondents believed that their group’s AI Governance program was inadequate in guaranteeing the protection and compliance of their AI property and use circumstances. As well as, knowledge privateness and safety breaches had been the highest concern for 53% of enterprise architects, whereas safety and governance are probably the most difficult points of knowledge engineering for engineers.
As well as, in line with Gartner, AI belief, danger, and safety administration is the #1 high technique pattern in 2024 that may issue into enterprise and know-how choices, and by 2026, AI fashions from organizations that operationalize AI transparency, belief, and safety will obtain 50% enhance by way of adoption, enterprise targets, and person acceptance.
Whereas it’s evident that the dearth of enterprise-level AI governance applications is quick turning into a key blocker to realizing return on worth from AI investments and AI adoption as a complete, we realized that there’s not a single, complete steering framework that enterprises can leverage to construct efficient AI governance applications.
The 5 foundational pillars
On this framework, we introduce 43 key issues which can be important for each enterprise to grasp (and implement as acceptable) to successfully govern their AI journeys.
These key issues had been then logically grouped throughout 5 foundational pillars, designed and sequenced to replicate typical enterprise org-structures and personas.
Pillar I: AI Group
The AI Group pillar embeds AI governance inside the group’s broader governance technique. It underscores the inspiration for an efficient AI program by greatest practices like clearly outlined enterprise aims and integrating the suitable governance practices that oversee the group’s individuals, processes, know-how, and knowledge. It explains how organizations can set up the oversight required to attain their strategic targets whereas decreasing danger.
Pillar II: Authorized and Regulatory Compliance
The Authorized and Regulatory Compliance pillar helps organizations align AI initiatives with relevant legal guidelines and rules. It guides managing authorized dangers, deciphering sector-specific necessities, and adapting compliance methods in response to evolving regulatory landscapes. The result is AI applications are developed and deployed inside a sturdy authorized and regulatory framework.
Pillar III: Ethics, Transparency and Interpretability
The Ethics, Transparency, and Interpretability pillar helps organizations in constructing reliable and accountable AI techniques. It emphasizes adherence to moral ideas resembling equity, accountability, and human oversight whereas selling explainability and stakeholder engagement. This pillar gives strategies to determine accountability and construction inside organizational groups, serving to to make sure that AI choices are interpretable, aligned with evolving moral requirements, and fostering long-term belief and societal acceptance.
Pillar IV: Information, AI Ops, and Infrastructure
The Information, AI Operations (AIOps), and Infrastructure pillar defines the inspiration that helps organizations in absolutely deploying and sustaining AI. It gives pointers for making a scalable and dependable AI infrastructure, managing the machine studying lifecycle, and guaranteeing knowledge high quality, safety, and compliance. This pillar additionally emphasizes greatest practices for AI operations, together with mannequin coaching, analysis, deployment, and monitoring, so AI techniques are dependable, environment friendly, and aligned with enterprise targets.
Pillar V: AI Safety
The AI Safety pillar introduces the Databricks AI Safety Framework (DASF), a complete framework for understanding and mitigating safety dangers throughout the AI lifecycle. It covers essential areas resembling knowledge safety, mannequin administration, safe mannequin serving, and the implementation of strong cybersecurity measures to guard AI property.
For an extra overview of DAGF and for an instance walkthrough of how a company can leverage the framework to create clear possession and alignment throughout the AI program lifecycle, please watch this presentation from the authors made throughout the 2025 Information + AI Summit.
Why Databricks is main this effort
As an business chief within the knowledge and AI house, with over 15,000 prospects throughout various geographies and market segments, Databricks has continued to ship on its dedication to ideas of accountable improvement and open supply innovation. We’ve upheld these commitments by our:
- Engagement with each business and authorities efforts to advertise innovation and advocate for using protected and reliable AI
- Interactive workshops to coach organizations on learn how to efficiently shepherd their AI journey in a risk-conscious method
- Open sourcing of key governance improvements resembling MLFlow and Unity Catalog, the business’s solely unified answer for knowledge and AI governance throughout clouds, knowledge codecs and knowledge platforms.
These applications have provided us distinctive visibility into sensible issues that enterprises and regulators face at the moment in AI governance. In furthering our dedication to serving to each enterprise succeed and speed up their Information and AI journey, we determined to leverage this visibility to construct (and make freely obtainable) a complete, structured and actionable AI Governance Framework.
Obtain the Databricks AI Governance Framework at the moment!
The Databricks AI Governance Framework whitepaper is now obtainable for obtain. Please attain out to us by way of e mail at [email protected] for any questions or suggestions. When you’re excited about contributing to future updates of this framework (and different upcoming artifacts) by becoming a member of our reviewer group, we’d love to listen to from you as effectively!
