In the present day, we’re introducing the Databricks AI Governance Framework (DAGF v1.0), a structured and sensible strategy 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 help program growth, deployment, and steady enchancment.
The DAGF enhances the Databricks AI Safety Framework, providing an entire view of governance that spans each safety and operational integrity.
Why AI governance can’t wait
In accordance with a 2024 international survey of 1,100 expertise executives and engineers carried out by Economist Influence, 40% of respondents believed that their group’s AI Governance program was inadequate in guaranteeing the protection and compliance of their AI belongings and use circumstances. 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 points of knowledge engineering for engineers.
As well as, in accordance with Gartner, AI belief, danger, and safety administration is the #1 prime technique pattern in 2024 that can 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 consumer acceptance.
Whereas it’s evident that the dearth of enterprise-level AI governance applications is quick changing 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 are 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 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 means of greatest practices like clearly outlined enterprise aims 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 attain their strategic objectives whereas lowering 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, decoding sector-specific necessities, and adapting compliance methods in response to evolving regulatory landscapes. The result is AI applications are developed and deployed inside a strong 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 ideas akin to equity, accountability, and human oversight whereas selling explainability and stakeholder engagement. This pillar offers strategies to ascertain 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 absolutely deploying and sustaining AI. It offers tips for making a scalable and dependable AI infrastructure, managing the machine studying lifecycle, and guaranteeing information high quality, safety, and compliance. This pillar additionally emphasizes greatest 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 important areas akin to information safety, mannequin administration, safe mannequin serving, and the implementation of strong cybersecurity measures to guard AI belongings.
For a further 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 in the course of the 2025 Knowledge + AI Summit.
Why Databricks is main this effort
As an business 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 ideas of accountable growth and open supply innovation. We’ve upheld these commitments by means of our:
- Engagement with each business and authorities efforts to advertise innovation and advocate for the usage of secure and reliable AI
- Interactive workshops to coach organizations on easy methods to efficiently shepherd their AI journey in a risk-conscious method
- Open sourcing of key governance improvements akin to MLFlow and Unity Catalog, the business’s solely unified resolution for information and AI governance throughout clouds, information codecs and information platforms.
These applications have provided us distinctive visibility into sensible issues that enterprises and regulators face as we speak 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 obtainable) a complete, structured and actionable AI Governance Framework.
Obtain the Databricks AI Governance Framework as we speak!
The Databricks AI Governance Framework whitepaper is now obtainable for obtain. Please attain out to us through electronic mail at [email protected] for any questions or suggestions. In the event you’re involved in 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!
