As artificial intelligence systems rapidly evolve and start to impact nearly every sector of society, the conversation around governance has mainly focused on models (and their output): their transparency, fairness, accountability, and alignment. Yet this focus, while necessary, is incomplete. AI systems are only as reliable, equitable, and effective as the data (input) on which they are trained and operate.

Data governance is not peripheral to AI governance — it is its bedrock.

At the same time, the rise of AI is not simply placing new demands on data governance; it is fundamentally transforming it. What counts as data, how it is curated, who has a say in its use, and which institutional arrangements govern it are all being reimagined in response to AI’s capabilities and risks.

This essay examines 10 key areas or shifts where data governance is being reshaped—either to accommodate AI or as a direct consequence of it.

1. Redefining What Counts as Data

Historically, data governance focused on structured, tabular datasets—administrative records, surveys, and spreadsheets. Today, the center of gravity has shifted toward unstructured data: text, images, audio, video, and multimodal content that fuel large-scale models.

Large Language Models (LLMs) in particular rely on vast corpora often scraped from the web, raising new governance challenges around provenance, consent, copyright, and representativeness.

At the same time, AI is no longer just a consumer of data — it is a producer of data. Synthetic text, images, and signals generated by AI systems are increasingly fed back into training pipelines, raising the specter of model collapse and necessitating governance frameworks for machine-generated data itself.

2. From FAIR to FAIR-R

The FAIR principles — Findable, Accessible, Interoperable, Reusable — have long guided data stewardship and remain fundamental to enabling responsible access to data for reuse. But AI systems require more.

We are witnessing the need to extend FAIR to FAIR-R (Ready-for-AI) to include the following:

• Structured metadata for machine interpretability;
• Better documentation of lineage and provenance;
• Bias and representativeness assessments;
• Alignment with responsible AI practices.

In short, data must now be not only reusable but also reusable by machines in ways that are safe, auditable, and aligned with societal values.

3. The Rise of Context as Infrastructure

Data without context is increasingly unusable in AI systems. Models require not just raw inputs but structured information about meaning, relationships, and intended use.

This has led to the development of new protocols such as Model Context Protocol (MCP), which aims to standardize how context—such as tools, memory, and environmental information—is structured and transmitted to AI systems alongside data.

Context is becoming a form of infrastructure — governed, curated, and standardized — shaping how AI systems interpret and act upon data.

4. From Data Stewardship to Strategic Data Stewardship

Traditional data stewardship focused on compliance, management, and quality control. In the age of AI, this is insufficient.

There is a growing need for strategic data stewardship — a proactive, purpose-driven approach that:

• Aligns data use with public value
• Anticipates downstream AI applications
• Brokers relationships across sectors
• Enables responsible reuse at scale

The role of the data steward is evolving from custodian to orchestrator of data ecosystems.

5. New Licensing Regimes for the AI Era

Existing data licensing frameworks—such as Creative Commons—were not designed with AI training in mind.

In response, new approaches are emerging, including AI-specific licenses and signaling mechanisms (e.g., “cc-signal”) that indicate whether and how data can be used for model training.

These developments reflect a broader transition: from static permissioning toward signalling of preferences and more dynamic, machine-readable governance of data rights.

6. Social License and Participatory Governance

Legal compliance and acquiring consent alone are no longer sufficient to legitimize data use in AI systems. Public trust increasingly depends on broader notions of social license.

We are seeing the rise of participatory mechanisms—citizen assemblies, stakeholder consultations, community governance models—that allow affected groups to shape decisions about how their data is used.

This marks a shift from consent as a transaction to agency as a process.

7. New Institutional Forms: Data Commons and Beyond

To counter risks of data extraction and concentration, new institutional arrangements are emerging, including data commons, cooperatives, and trusts.

These models aim to:

• Embed collective governance;
• Align data use with community preferences and shared goals;
• Redistribute value generated from data.

In the AI context, such arrangements are critical to ensuring that data is not merely extracted but mobilized for collective agency and public benefit.

8. Synthetic Data as a Governance Tool

Synthetic data—artificially generated datasets that mimic real-world patterns—has gained traction as a way to address privacy, access, and scarcity challenges.

When done responsibly, its governance implications include:

• It can enable safe data sharing without exposing sensitive information;
• It can fill gaps in underrepresented datasets;
• It raises questions about fidelity, bias amplification, and misuse.

As such, synthetic data is not just a technical solution — it is a new object of governance in its own right.

9. AI for Data Governance

AI is not only governed by data governance — it is increasingly used to perform data governance. Applications include:

• Automated data discovery and classification;
• Quality assessment and anomaly detection;
• Monitoring compliance and usage patterns;
• Auditing datasets and models for bias and risk.

This introduces both efficiencies and new risks, as governance itself becomes partially automated.

10. The Emergence of AI Agents in Data Governance

Finally, the rise of AI agents — systems capable of autonomous, multi-step decision-making — signals a new frontier as it relates to data management and governance. These agents have been used to:

• Negotiate data access
• Enforce governance rules
• Manage data pipelines dynamically
• Act as intermediaries between users and data ecosystems

This raises fundamental questions about delegation, accountability, and control in governance systems where machines act on behalf of humans.

Conclusion: Data Governance as a Living Practice

Data governance is the foundation upon which AI systems are built. But in the age of AI, it is no longer a static foundation — it is a dynamic, evolving practice and system shaped by the very technologies it enables.

We are moving toward a world where:

• Data governance shapes AI;
• AI reshapes data governance;
• And both co-evolve in a continuous feedback loop.

The challenge ahead is not simply to adapt existing frameworks but to reimagine data governance as a living practice and system—capable of ensuring that AI serves not only efficiency and innovation but also equity, accountability, and the public good.

Image C: Daniela Zampieri / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/