Over the last several months, The Open Data Policy Lab conducted an in-depth review of how the data commons gathered within our Data Commons for Generative AI repository align with the six elements of our data commons assessment framework: Public Purpose, Contributor Benefits, Funding and Support, Access Controls, Participatory Governance, and Accountability Mechanisms. 

The review revealed both patterns and variation. Many initiatives articulate a public purpose, offer benefits to contributors, identify funding or support, and define access conditions. Participatory governance and accountability mechanisms varied more widely and often marked the clearest distinctions between data commons and adjacent models such as open data platforms, data cooperatives, and AI data pools.

Our review highlighted that data commons take many institutional forms, including scientific research infrastructures, health data collaboratives, cultural and linguistic preservation efforts, industrial data spaces, and community-owned initiatives. Some operate globally or nationally, while others focus on specific sectors, communities, or local challenges.

In the sections that follow, we examine what the repository reveals through each element of our framework, offering a snapshot of how organizations across sectors are attempting to build, govern, and sustain shared data resources in an era shaped by data-intensive technologies and artificial intelligence.

Key Findings Across Our Six Elements of Data Commons 

Public Purpose

Public purpose generally refers to the commitment to transform data into a shared resource in the public’s interest. The initiatives in our repository formalized this commitment through missions, governance structures, or stewardship arrangements designed to align data use with broader societal goals. 

Several initiatives in our repository focus on scientific and clinical advancement. Initiatives such as MIDRC, the NCI Cancer Research Data Commons, INSIGHT, and AIDA Data Hub seek to improve disease detection, diagnosis, treatment, and care delivery by enabling researchers to access high-quality datasets that would otherwise remain siloed within individual hospitals, laboratories, or research institutions. In these cases, public purpose is closely tied to patient benefit, improved health outcomes, and the responsible reuse of sensitive biomedical data.

Many initiatives identified seek to improve public-interest AI innovation, fairness, and inclusion. MLCommons, Common Voice, WAXAL, and the Malawi Voice Data Commons seek to expand access to representative, high-quality data so that AI development does not systematically exclude underrepresented languages, regions, or communities.

Others aim to support cultural and linguistic preservation and revitalization. The Language Data Commons of Australia, CLARIN, Open Heritage, MetaBelgica, and the Common European Data Space for Cultural Heritage treat data as a vehicle for preserving cultural memory, linguistic diversity, and shared knowledge. These initiatives emphasize long-term stewardship, responsible access, cultural authority, and the preservation of resources that might otherwise disappear.

We also identified initiatives that pursue public purpose through democratic values, environmental resilience, and community self-determination. For example, the Public Interest Corpus seeks to support public-interest AI research and a more informed knowledge ecosystem. Posmo advances informational self-determination by giving mobility-data contributors greater control over how their data is used. GainForest.Earth supports local and Indigenous communities in collecting and governing biodiversity and environmental data. Sectoral commons such as Catena-X and the Mobility Data Space pursue public value through industry-wide coordination, sustainability, interoperability, and supply-chain transparency.

The repository shows that public purpose is not a single objective but a spectrum of social commitments. Some initiatives define the public in broad societal terms, while others focus on specific communities, sectors, or domains. 

Contributor Benefits

Throughout the repository, contributors are treated as stakeholders whose continued participation depends on tangible value. Like public purpose, contributor benefits appear in nearly all initiatives in the repository. The few exceptions are typically initiatives in development, where benefits are anticipated but not yet fully operationalized.

The most prevalent contributor benefit is visibility and academic credit. Initiatives such as AIDA Data Hub make datasets more FAIR and citable by assigning Digital Object Identifiers, allowing researchers to receive formal recognition when their datasets are reused. The Health Data Hub similarly reinforces contributor visibility by requiring projects to cite source databases and by informing contributors when their data is reused.

Several initiatives support their contributors through technical and operational support. Some initiatives reduce the burdens of data management by helping contributors with extraction, documentation, harmonization, secure access workflows, and infrastructure. The Mobility Data Space, for example, provides a Connector-as-a-Service model that allows organizations to exchange data securely without maintaining complex technical infrastructure themselves. The HathiTrust Research Center offers members specialized developer support, technical expertise, and high-capacity computing environments for large-scale text and data mining. 

Another benefit is influence over standards, strategy, and governance. Catena-X gives participating organizations opportunities to co-create industry standards, define API specifications, and shape certification requirements for the automotive ecosystem. In the Posmo cooperative, individuals who contribute mobility data gain membership rights and can participate directly in decisions about how the data is used and which institutions may access it. These models treat governance itself as a contributor benefit by giving participants influence over the future direction of the commons. 

Some initiatives provide financial incentives or data enrichment. AIDA Data Hub offers discounts and other incentives for contributors of prioritized datasets while also supporting researchers to extract, enrich, and prepare data for broader sharing and reuse. GainForest.Earth rewards communities for contributing biodiversity data through community-defined pricing models while also providing training, AI tools, and conservation technologies that strengthen local capacity to collect, steward, and benefit from environmental data. The Health Data Hub offers another type of benefit, linking and enriching contributed datasets with additional health data resources, allowing contributors to derive greater value from the data they originally supplied.

Funding and Support

Many initiatives within our repository begin with grants, public seed funding, or pilot support, but durable commons require longer-term arrangements to cover the ongoing costs of infrastructure, stewardship, governance, accountability, and community engagement. As we have discussed elsewhere, funding models frequently overlap and evolve over time.

The majority of initiatives are supported by multiple organizations through consortium or collaborative arrangements. Catena-X, MLCommons, CLARIN, and Open Heritage bring together public agencies, research institutions, companies, philanthropies, and civil society organizations that contribute funding, infrastructure, technical expertise, governance labor, or domain knowledge.

A second major theme is public or government-backed funding. Health Data Hub, INSIGHT, the NCI Cancer Research Data Commons, Mobility Data Space, and the European Open Science Cloud depend on funding from government agencies, research councils, public institutions, or state-supported programs. 

Other initiatives sustain themselves through membership dues, service models, or donations. HathiTrust, AIDA Data Hub, OpenStreetMap, and parts of the Wikimedia ecosystem rely on combinations of institutional memberships, service fees, philanthropic donations, and volunteer contributions. Initiatives such as Common Voice, the Public Interest Corpus, TRANSFER Data Trust, GainForest.Earth, and the Malawi Voice Data Commons rely on philanthropic, grant-based, or early-stage development funding. These initiatives depend on grants, fiscal sponsorship, foundation support, or pilot funding to address public-interest challenges such as democratizing AI, strengthening community data governance, preserving language and cultural resources, or expanding access to scientific and educational data.

Access Controls

Across the repository, access is rarely treated as simply “open” or “closed.” Instead, initiatives use legal, procedural, technical, and community-based mechanisms to balance data availability with privacy, consent, intellectual property, cultural authority, commercial sensitivity, and data sovereignty. 

Several initiatives use ethical review processes to determine whether data access should be granted. INSIGHT, for example, uses the Five Safes framework and routes feasible applications to its Data Trust Advisory Board, which includes patient and public stakeholders and evaluates whether proposed uses balance scientific discovery with public benefit. The Health Data Hub follows a similarly formal process, requiring scientific and ethical review before authorization by France’s national data protection authority.

A second model is secure analysis, where users work with data inside a controlled environment. The HathiTrust Research Center uses Data Capsules to support computational research on protected collections while preventing substantial portions of raw texts from leaving the system; exports must be reviewed to ensure they do not reproduce protected expressive content. INSIGHT similarly provides access through a Secure Research Environment, where researchers work in controlled virtual environments and outputs are reviewed before export. 

Other initiatives rely on credentials, training, and data use agreements to ensure that only qualified or accountable users gain access. MIDRC, for example, requires user registration, institutional authentication, and agreement to detailed data use conditions governing redistribution, re-identification, publication, and commercial use. Health Data Nexus similarly combines institutional authorization, mandatory ethics training, and data use agreements before granting access to specific datasets. 

For lower-risk or more openly shared resources, access controls often take the form of standardized licenses, APIs, and tiered access categories. CLARIN organizes language resources into public, academic, and restricted-access categories, while initiatives such as Common Voice and Institutional Books 1.0 rely on open licenses, attribution requirements, platform terms, or API-based access. These models maximize reuse while still establishing expectations for lawful and responsible use.

Industrial and federated data spaces use access controls rooted in data sovereignty. Mobility Data Space and Catena-X allow data to remain with providers while exchange is mediated through connectors, contracts, and shared technical standards. 

Participatory Governance

Participatory governance is the primary feature that distinguishes a data commons from a platform for data access. Throughout the repository, many initiatives involved consultation, contribution, or co-design, but only some gave participants authority over access conditions, technical standards, strategic priorities, or future development. In some commons, individuals or communities participate directly; in others, participation occurs through organizations, institutions, or sectoral representatives. What matters is not who participates, but whether decision-making authority is meaningfully shared.

One governance model is independent stakeholder oversight, where affected groups or public representatives help determine whether data use is legitimate. INSIGHT gives its Data Trust Advisory Board authority over the assessment criteria used to approve or deny data access requests. The Health Data Hub follows a more institutional version of this model through CESREES—the Ethical and Scientific Committee for Research, Studies and Evaluations in the field of Health—which evaluates the scientific relevance, ethical character, and public-interest value of proposed projects.

A second model is collaborative standard-setting, where participating organizations collectively shape the technical and operational rules of the commons. This approach is common in consortium-based initiatives where membership is held by institutions. In Catena-X, for example, members participate in topic-specific working groups that define standards, data models, API specifications, and software priorities for the automotive ecosystem. 

A third model centers on community sovereignty and member control. Posmo gives every individual who contributes data the right to become a member and participate in decisions about which institutions may access aggregated datasets. TRANSFER Data Trust similarly uses a member-owned cooperative model that gives artists and cultural organizations a direct role in shaping stewardship and governance. The Language Data Commons of Australia extends this logic to Indigenous and community authority by using governance structures informed by CARE principles to support appropriate control over sensitive linguistic collections.

Other initiatives rely on strategic steering and advisory bodies to guide priorities. Artsdata, for example, is governed by the Artsdata Community Group, which brings together data providers, users, financial supporters, and sector experts to guide the governance and evolution of the shared knowledge graph.

Like the other elements, participatory governance does not take a single institutional form. It can involve patient and public oversight, collaborative ecosystem governance, cooperative ownership, Indigenous and community data sovereignty, advisory councils, or technical standard-setting. It may operate through individual contributors, community representatives, or participating organizations. What distinguishes a data commons is the distribution of authority: meaningful decisions about access, stewardship, standards, and future development are shared rather than retained exclusively by a central actor.

Accountability Mechanisms

Accountability mechanisms clarify who is responsible for data use, how compliance is monitored, and how contributors, communities, or the public can assess whether the commons is being used appropriately.

Several initiatives use public transparency and usage reporting. INSIGHT’s Data Use Register provides an accessible record of the project title, lead applicant, and purpose for each study granted access to NHS patient data. The Health Data Hub similarly publishes information about datasets in its catalogue and the research projects using them. The Public Interest Corpus also emphasizes transparency around corpus composition and future modifications as a way to build trust in downstream research.

A second model is institutional and professional responsibility. AIDA Data Hub requires access requests to be led by a qualified researcher affiliated with a responsible research institution, which remains accountable for safeguarding the data according to its policies and applicable law. At the Health Data Nexus, users must sign a Data Use Agreement for each dataset they wish to access, linking their institutional identity to specific ethical and legal obligations.

A third model relies on technical auditing and traceability. In industrial data spaces such as the Mobility Data Space, accountability is embedded in usage contracts between data providers and recipients and supported through technical infrastructure. The Health Data Hub uses a secure platform that traces actions performed on data, creating an audit trail for compliance. Within INSIGHT’s Secure Research Environment, a data custodian must approve the export of analysis results to ensure that sensitive information does not leave the protected environment.

Other initiatives ground accountability in ethical frameworks, standards, and training. The Language Data Commons of Australia draws on CARE principles to support culturally appropriate stewardship of sensitive language data. INSIGHT uses the Five Safes framework to assess whether projects involve safe people, safe data, safe settings, safe outputs, and safe projects. 

Reflections

Applying the assessment framework clarified that data commons are not simply repositories, platforms, or datasets, but socio-technical institutions. They must balance the needs of data providers, users, data subjects, affected communities, funders, and the broader public interest while sustaining the rules, relationships, and infrastructure that make responsible data access possible. 

First, trust is essential. Across the repository, initiatives used access controls, accountability mechanisms, and ethical review to enable data access, especially where data is sensitive, culturally significant, commercially valuable, or personally identifiable. Secure research environments, public data-use registers, ethical review boards, contributor-controlled permissions, and traceable audit systems help create the conditions under which communities and institutions are willing to provide access.

Second, data commons cannot depend on altruism alone. They require reciprocity. Contributors need meaningful value in return for the labor, risk, and expertise involved in preparing and sharing data. That value may come through access to a larger and better-curated shared resource, attribution, persistent identifiers, technical support, data enrichment, capacity-building, governance rights, or strategic influence. 

Third, grants, philanthropic support, public seed funding, and time-limited research funding can help launch initiatives, but they do not guarantee long-term sustainability. More durable initiatives combine funding sources such as public mandates, institutional support, membership dues, service fees, donations, volunteer labor, and consortium contributions. 

Finally, governance is the clearest boundary between a data commons and other forms of collaborative data sharing. Open datasets, data sharing platforms, standards bodies, and commons-enabling alliances can all contribute to the broader data ecosystem, but they do not automatically qualify as data commons. What matters is whether contributors, affected communities, members, or participating institutions hold meaningful authority over access rules, standards, stewardship practices, accountability mechanisms, or future development. This shared authority separates participation in building a data resource from participation in governing it as a commons. 

***

To learn more about our broader data commons work, including the New Commons Incubator, visit newcommons.ai.  

Back to the Blog