Right now, 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 growth with enterprise objectives, meet authorized obligations, and account for moral dangers. This framework is designed to assist program growth, 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 line with a 2024 international survey of 1,100 expertise executives and engineers carried out by Economist Affect, 40% of respondents believed that their group’s AI Governance program was inadequate in making certain the protection and compliance of their AI property and use instances. As well as, information privateness and safety breaches had been the highest concern for 53% of enterprise architects, whereas safety and governance are essentially the most difficult elements of information engineering for engineers.
As well as, in accordance with Gartner, AI belief, threat, and safety administration is the #1 high technique pattern in 2024 that may issue into enterprise and expertise selections, and by 2026, AI fashions from organizations that operationalize AI transparency, belief, and safety will obtain 50% enhance by way of adoption, enterprise objectives, and person acceptance.
Whereas it’s evident that the shortage 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 an entire, we realized that there’s not a single, complete steerage 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 applicable) to successfully govern their AI journeys.
These key issues had been then logically grouped throughout 5 foundational pillars, designed and sequenced to mirror typical enterprise org-structures and personas.
Pillar I: AI Group
The AI Group pillar embeds AI governance throughout the group’s broader governance technique. It underscores the muse for an efficient AI program by way of finest practices like clearly outlined enterprise targets and integrating the suitable governance practices that oversee the group’s individuals, processes, expertise, and information. It explains how organizations can set up the oversight required to realize their strategic objectives whereas decreasing threat.
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 methods. It emphasizes adherence to moral rules comparable to 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 selections are interpretable, aligned with evolving moral requirements, and fostering long-term belief and societal acceptance.
Pillar IV: Knowledge, AI Ops, and Infrastructure
The Knowledge, AI Operations (AIOps), and Infrastructure pillar defines the muse that helps organizations in totally deploying and sustaining AI. It gives tips for making a scalable and dependable AI infrastructure, managing the machine studying lifecycle, and making certain information high quality, safety, and compliance. This pillar additionally emphasizes finest practices for AI operations, together with mannequin coaching, analysis, deployment, and monitoring, so AI methods are dependable, environment friendly, and aligned with enterprise objectives.
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 crucial areas comparable to information safety, mannequin administration, safe mannequin serving, and the implementation of sturdy cybersecurity measures to guard AI property.
For a further overview of DAGF and for an instance walkthrough of how a corporation can leverage the framework to create clear possession and alignment throughout the AI program lifecycle, please watch this presentation from the authors made through the 2025 Knowledge + AI Summit.
Why Databricks is main this effort
As an trade chief within the information and AI area, with over 15,000 clients throughout various geographies and market segments, Databricks has continued to ship on its dedication to rules of accountable growth and open supply innovation. We’ve upheld these commitments by way of our:
- Engagement with each trade and authorities efforts to advertise innovation and advocate for the usage of protected and reliable AI
- Interactive workshops to teach organizations on find out how to efficiently shepherd their AI journey in a risk-conscious method
- Open sourcing of key governance improvements comparable to MLFlow and Unity Catalog, the trade’s solely unified answer for information and AI governance throughout clouds, information codecs and information platforms.
These applications have supplied us distinctive visibility into sensible issues that enterprises and regulators face right this moment in AI governance. In furthering our dedication to serving to each enterprise succeed and speed up their Knowledge and AI journey, we determined to leverage this visibility to construct (and make freely accessible) a complete, structured and actionable AI Governance Framework.
Obtain the Databricks AI Governance Framework right this moment!
The Databricks AI Governance Framework whitepaper is now accessible for obtain. Please attain out to us by way of e-mail at [email protected] for any questions or suggestions. In the event you’re desirous about contributing to future updates of this framework (and different upcoming artifacts) by becoming a member of our reviewer neighborhood, we’d love to listen to from you as properly!